Merge pull request #1134 from yasu-dev221/fix/openai-compat-prompt-cache-key

fix(openai): add fallback prompt_cache_key for compat codex OAuth requests

.github/workflows/release.yml

@@ -271,3 +271,36 @@ jobs:
                 parse_mode: "Markdown",
                 disable_web_page_preview: true
               }')"
+
+  sync-version-file:
+    needs: [release]
+    if: ${{ needs.release.result == 'success' }}
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout default branch
+        uses: actions/checkout@v6
+        with:
+          ref: ${{ github.event.repository.default_branch }}
+
+      - name: Sync VERSION file to released tag
+        run: |
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            VERSION=${{ github.event.inputs.tag }}
+            VERSION=${VERSION#v}
+          else
+            VERSION=${GITHUB_REF#refs/tags/v}
+          fi
+
+          CURRENT_VERSION=$(tr -d '\r\n' < backend/cmd/server/VERSION || true)
+          if [ "$CURRENT_VERSION" = "$VERSION" ]; then
+            echo "VERSION file already matches $VERSION"
+            exit 0
+          fi
+
+          echo "$VERSION" > backend/cmd/server/VERSION
+
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          git add backend/cmd/server/VERSION
+          git commit -m "chore: sync VERSION to ${VERSION} [skip ci]"
+          git push origin HEAD:${{ github.event.repository.default_branch }}

backend/ent/migrate/schema.go

@@ -716,6 +716,7 @@ var (
 		{Name: "id", Type: field.TypeInt64, Increment: true},
 		{Name: "request_id", Type: field.TypeString, Size: 64},
 		{Name: "model", Type: field.TypeString, Size: 100},
+		{Name: "upstream_model", Type: field.TypeString, Nullable: true, Size: 100},
 		{Name: "input_tokens", Type: field.TypeInt, Default: 0},
 		{Name: "output_tokens", Type: field.TypeInt, Default: 0},
 		{Name: "cache_creation_tokens", Type: field.TypeInt, Default: 0},
@@ -755,31 +756,31 @@ var (
 		ForeignKeys: []*schema.ForeignKey{
 			{
 				Symbol:     "usage_logs_api_keys_usage_logs",
-				Columns:    []*schema.Column{UsageLogsColumns[28]},
+				Columns:    []*schema.Column{UsageLogsColumns[29]},
 				RefColumns: []*schema.Column{APIKeysColumns[0]},
 				OnDelete:   schema.NoAction,
 			},
 			{
 				Symbol:     "usage_logs_accounts_usage_logs",
-				Columns:    []*schema.Column{UsageLogsColumns[29]},
+				Columns:    []*schema.Column{UsageLogsColumns[30]},
 				RefColumns: []*schema.Column{AccountsColumns[0]},
 				OnDelete:   schema.NoAction,
 			},
 			{
 				Symbol:     "usage_logs_groups_usage_logs",
-				Columns:    []*schema.Column{UsageLogsColumns[30]},
+				Columns:    []*schema.Column{UsageLogsColumns[31]},
 				RefColumns: []*schema.Column{GroupsColumns[0]},
 				OnDelete:   schema.SetNull,
 			},
 			{
 				Symbol:     "usage_logs_users_usage_logs",
-				Columns:    []*schema.Column{UsageLogsColumns[31]},
+				Columns:    []*schema.Column{UsageLogsColumns[32]},
 				RefColumns: []*schema.Column{UsersColumns[0]},
 				OnDelete:   schema.NoAction,
 			},
 			{
 				Symbol:     "usage_logs_user_subscriptions_usage_logs",
-				Columns:    []*schema.Column{UsageLogsColumns[32]},
+				Columns:    []*schema.Column{UsageLogsColumns[33]},
 				RefColumns: []*schema.Column{UserSubscriptionsColumns[0]},
 				OnDelete:   schema.SetNull,
 			},
@@ -788,32 +789,32 @@ var (
 			{
 				Name:    "usagelog_user_id",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[31]},
+				Columns: []*schema.Column{UsageLogsColumns[32]},
 			},
 			{
 				Name:    "usagelog_api_key_id",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[28]},
+				Columns: []*schema.Column{UsageLogsColumns[29]},
 			},
 			{
 				Name:    "usagelog_account_id",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[29]},
+				Columns: []*schema.Column{UsageLogsColumns[30]},
 			},
 			{
 				Name:    "usagelog_group_id",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[30]},
+				Columns: []*schema.Column{UsageLogsColumns[31]},
 			},
 			{
 				Name:    "usagelog_subscription_id",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[32]},
+				Columns: []*schema.Column{UsageLogsColumns[33]},
 			},
 			{
 				Name:    "usagelog_created_at",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[27]},
+				Columns: []*schema.Column{UsageLogsColumns[28]},
 			},
 			{
 				Name:    "usagelog_model",
@@ -828,17 +829,17 @@ var (
 			{
 				Name:    "usagelog_user_id_created_at",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[31], UsageLogsColumns[27]},
+				Columns: []*schema.Column{UsageLogsColumns[32], UsageLogsColumns[28]},
 			},
 			{
 				Name:    "usagelog_api_key_id_created_at",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[28], UsageLogsColumns[27]},
+				Columns: []*schema.Column{UsageLogsColumns[29], UsageLogsColumns[28]},
 			},
 			{
 				Name:    "usagelog_group_id_created_at",
 				Unique:  false,
-				Columns: []*schema.Column{UsageLogsColumns[30], UsageLogsColumns[27]},
+				Columns: []*schema.Column{UsageLogsColumns[31], UsageLogsColumns[28]},
 			},
 		},
 	}

backend/ent/mutation.go

@@ -18239,6 +18239,7 @@ type UsageLogMutation struct {
 	id                          *int64
 	request_id                  *string
 	model                       *string
+	upstream_model              *string
 	input_tokens                *int
 	addinput_tokens             *int
 	output_tokens               *int
@@ -18576,6 +18577,55 @@ func (m *UsageLogMutation) ResetModel() {
 	m.model = nil
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (m *UsageLogMutation) SetUpstreamModel(s string) {
+	m.upstream_model = &s
+}
+
+// UpstreamModel returns the value of the "upstream_model" field in the mutation.
+func (m *UsageLogMutation) UpstreamModel() (r string, exists bool) {
+	v := m.upstream_model
+	if v == nil {
+		return
+	}
+	return *v, true
+}
+
+// OldUpstreamModel returns the old "upstream_model" field's value of the UsageLog entity.
+// If the UsageLog object wasn't provided to the builder, the object is fetched from the database.
+// An error is returned if the mutation operation is not UpdateOne, or the database query fails.
+func (m *UsageLogMutation) OldUpstreamModel(ctx context.Context) (v *string, err error) {
+	if !m.op.Is(OpUpdateOne) {
+		return v, errors.New("OldUpstreamModel is only allowed on UpdateOne operations")
+	}
+	if m.id == nil || m.oldValue == nil {
+		return v, errors.New("OldUpstreamModel requires an ID field in the mutation")
+	}
+	oldValue, err := m.oldValue(ctx)
+	if err != nil {
+		return v, fmt.Errorf("querying old value for OldUpstreamModel: %w", err)
+	}
+	return oldValue.UpstreamModel, nil
+}
+
+// ClearUpstreamModel clears the value of the "upstream_model" field.
+func (m *UsageLogMutation) ClearUpstreamModel() {
+	m.upstream_model = nil
+	m.clearedFields[usagelog.FieldUpstreamModel] = struct{}{}
+}
+
+// UpstreamModelCleared returns if the "upstream_model" field was cleared in this mutation.
+func (m *UsageLogMutation) UpstreamModelCleared() bool {
+	_, ok := m.clearedFields[usagelog.FieldUpstreamModel]
+	return ok
+}
+
+// ResetUpstreamModel resets all changes to the "upstream_model" field.
+func (m *UsageLogMutation) ResetUpstreamModel() {
+	m.upstream_model = nil
+	delete(m.clearedFields, usagelog.FieldUpstreamModel)
+}
+
 // SetGroupID sets the "group_id" field.
 func (m *UsageLogMutation) SetGroupID(i int64) {
 	m.group = &i
@@ -20197,7 +20247,7 @@ func (m *UsageLogMutation) Type() string {
 // order to get all numeric fields that were incremented/decremented, call
 // AddedFields().
 func (m *UsageLogMutation) Fields() []string {
-	fields := make([]string, 0, 32)
+	fields := make([]string, 0, 33)
 	if m.user != nil {
 		fields = append(fields, usagelog.FieldUserID)
 	}
@@ -20213,6 +20263,9 @@ func (m *UsageLogMutation) Fields() []string {
 	if m.model != nil {
 		fields = append(fields, usagelog.FieldModel)
 	}
+	if m.upstream_model != nil {
+		fields = append(fields, usagelog.FieldUpstreamModel)
+	}
 	if m.group != nil {
 		fields = append(fields, usagelog.FieldGroupID)
 	}
@@ -20312,6 +20365,8 @@ func (m *UsageLogMutation) Field(name string) (ent.Value, bool) {
 		return m.RequestID()
 	case usagelog.FieldModel:
 		return m.Model()
+	case usagelog.FieldUpstreamModel:
+		return m.UpstreamModel()
 	case usagelog.FieldGroupID:
 		return m.GroupID()
 	case usagelog.FieldSubscriptionID:
@@ -20385,6 +20440,8 @@ func (m *UsageLogMutation) OldField(ctx context.Context, name string) (ent.Value
 		return m.OldRequestID(ctx)
 	case usagelog.FieldModel:
 		return m.OldModel(ctx)
+	case usagelog.FieldUpstreamModel:
+		return m.OldUpstreamModel(ctx)
 	case usagelog.FieldGroupID:
 		return m.OldGroupID(ctx)
 	case usagelog.FieldSubscriptionID:
@@ -20483,6 +20540,13 @@ func (m *UsageLogMutation) SetField(name string, value ent.Value) error {
 		}
 		m.SetModel(v)
 		return nil
+	case usagelog.FieldUpstreamModel:
+		v, ok := value.(string)
+		if !ok {
+			return fmt.Errorf("unexpected type %T for field %s", value, name)
+		}
+		m.SetUpstreamModel(v)
+		return nil
 	case usagelog.FieldGroupID:
 		v, ok := value.(int64)
 		if !ok {
@@ -20921,6 +20985,9 @@ func (m *UsageLogMutation) AddField(name string, value ent.Value) error {
 // mutation.
 func (m *UsageLogMutation) ClearedFields() []string {
 	var fields []string
+	if m.FieldCleared(usagelog.FieldUpstreamModel) {
+		fields = append(fields, usagelog.FieldUpstreamModel)
+	}
 	if m.FieldCleared(usagelog.FieldGroupID) {
 		fields = append(fields, usagelog.FieldGroupID)
 	}
@@ -20962,6 +21029,9 @@ func (m *UsageLogMutation) FieldCleared(name string) bool {
 // error if the field is not defined in the schema.
 func (m *UsageLogMutation) ClearField(name string) error {
 	switch name {
+	case usagelog.FieldUpstreamModel:
+		m.ClearUpstreamModel()
+		return nil
 	case usagelog.FieldGroupID:
 		m.ClearGroupID()
 		return nil
@@ -21012,6 +21082,9 @@ func (m *UsageLogMutation) ResetField(name string) error {
 	case usagelog.FieldModel:
 		m.ResetModel()
 		return nil
+	case usagelog.FieldUpstreamModel:
+		m.ResetUpstreamModel()
+		return nil
 	case usagelog.FieldGroupID:
 		m.ResetGroupID()
 		return nil

backend/ent/runtime/runtime.go

@@ -821,92 +821,96 @@ func init() {
 			return nil
 		}
 	}()
+	// usagelogDescUpstreamModel is the schema descriptor for upstream_model field.
+	usagelogDescUpstreamModel := usagelogFields[5].Descriptor()
+	// usagelog.UpstreamModelValidator is a validator for the "upstream_model" field. It is called by the builders before save.
+	usagelog.UpstreamModelValidator = usagelogDescUpstreamModel.Validators[0].(func(string) error)
 	// usagelogDescInputTokens is the schema descriptor for input_tokens field.
-	usagelogDescInputTokens := usagelogFields[7].Descriptor()
+	usagelogDescInputTokens := usagelogFields[8].Descriptor()
 	// usagelog.DefaultInputTokens holds the default value on creation for the input_tokens field.
 	usagelog.DefaultInputTokens = usagelogDescInputTokens.Default.(int)
 	// usagelogDescOutputTokens is the schema descriptor for output_tokens field.
-	usagelogDescOutputTokens := usagelogFields[8].Descriptor()
+	usagelogDescOutputTokens := usagelogFields[9].Descriptor()
 	// usagelog.DefaultOutputTokens holds the default value on creation for the output_tokens field.
 	usagelog.DefaultOutputTokens = usagelogDescOutputTokens.Default.(int)
 	// usagelogDescCacheCreationTokens is the schema descriptor for cache_creation_tokens field.
-	usagelogDescCacheCreationTokens := usagelogFields[9].Descriptor()
+	usagelogDescCacheCreationTokens := usagelogFields[10].Descriptor()
 	// usagelog.DefaultCacheCreationTokens holds the default value on creation for the cache_creation_tokens field.
 	usagelog.DefaultCacheCreationTokens = usagelogDescCacheCreationTokens.Default.(int)
 	// usagelogDescCacheReadTokens is the schema descriptor for cache_read_tokens field.
-	usagelogDescCacheReadTokens := usagelogFields[10].Descriptor()
+	usagelogDescCacheReadTokens := usagelogFields[11].Descriptor()
 	// usagelog.DefaultCacheReadTokens holds the default value on creation for the cache_read_tokens field.
 	usagelog.DefaultCacheReadTokens = usagelogDescCacheReadTokens.Default.(int)
 	// usagelogDescCacheCreation5mTokens is the schema descriptor for cache_creation_5m_tokens field.
-	usagelogDescCacheCreation5mTokens := usagelogFields[11].Descriptor()
+	usagelogDescCacheCreation5mTokens := usagelogFields[12].Descriptor()
 	// usagelog.DefaultCacheCreation5mTokens holds the default value on creation for the cache_creation_5m_tokens field.
 	usagelog.DefaultCacheCreation5mTokens = usagelogDescCacheCreation5mTokens.Default.(int)
 	// usagelogDescCacheCreation1hTokens is the schema descriptor for cache_creation_1h_tokens field.
-	usagelogDescCacheCreation1hTokens := usagelogFields[12].Descriptor()
+	usagelogDescCacheCreation1hTokens := usagelogFields[13].Descriptor()
 	// usagelog.DefaultCacheCreation1hTokens holds the default value on creation for the cache_creation_1h_tokens field.
 	usagelog.DefaultCacheCreation1hTokens = usagelogDescCacheCreation1hTokens.Default.(int)
 	// usagelogDescInputCost is the schema descriptor for input_cost field.
-	usagelogDescInputCost := usagelogFields[13].Descriptor()
+	usagelogDescInputCost := usagelogFields[14].Descriptor()
 	// usagelog.DefaultInputCost holds the default value on creation for the input_cost field.
 	usagelog.DefaultInputCost = usagelogDescInputCost.Default.(float64)
 	// usagelogDescOutputCost is the schema descriptor for output_cost field.
-	usagelogDescOutputCost := usagelogFields[14].Descriptor()
+	usagelogDescOutputCost := usagelogFields[15].Descriptor()
 	// usagelog.DefaultOutputCost holds the default value on creation for the output_cost field.
 	usagelog.DefaultOutputCost = usagelogDescOutputCost.Default.(float64)
 	// usagelogDescCacheCreationCost is the schema descriptor for cache_creation_cost field.
-	usagelogDescCacheCreationCost := usagelogFields[15].Descriptor()
+	usagelogDescCacheCreationCost := usagelogFields[16].Descriptor()
 	// usagelog.DefaultCacheCreationCost holds the default value on creation for the cache_creation_cost field.
 	usagelog.DefaultCacheCreationCost = usagelogDescCacheCreationCost.Default.(float64)
 	// usagelogDescCacheReadCost is the schema descriptor for cache_read_cost field.
-	usagelogDescCacheReadCost := usagelogFields[16].Descriptor()
+	usagelogDescCacheReadCost := usagelogFields[17].Descriptor()
 	// usagelog.DefaultCacheReadCost holds the default value on creation for the cache_read_cost field.
 	usagelog.DefaultCacheReadCost = usagelogDescCacheReadCost.Default.(float64)
 	// usagelogDescTotalCost is the schema descriptor for total_cost field.
-	usagelogDescTotalCost := usagelogFields[17].Descriptor()
+	usagelogDescTotalCost := usagelogFields[18].Descriptor()
 	// usagelog.DefaultTotalCost holds the default value on creation for the total_cost field.
 	usagelog.DefaultTotalCost = usagelogDescTotalCost.Default.(float64)
 	// usagelogDescActualCost is the schema descriptor for actual_cost field.
-	usagelogDescActualCost := usagelogFields[18].Descriptor()
+	usagelogDescActualCost := usagelogFields[19].Descriptor()
 	// usagelog.DefaultActualCost holds the default value on creation for the actual_cost field.
 	usagelog.DefaultActualCost = usagelogDescActualCost.Default.(float64)
 	// usagelogDescRateMultiplier is the schema descriptor for rate_multiplier field.
-	usagelogDescRateMultiplier := usagelogFields[19].Descriptor()
+	usagelogDescRateMultiplier := usagelogFields[20].Descriptor()
 	// usagelog.DefaultRateMultiplier holds the default value on creation for the rate_multiplier field.
 	usagelog.DefaultRateMultiplier = usagelogDescRateMultiplier.Default.(float64)
 	// usagelogDescBillingType is the schema descriptor for billing_type field.
-	usagelogDescBillingType := usagelogFields[21].Descriptor()
+	usagelogDescBillingType := usagelogFields[22].Descriptor()
 	// usagelog.DefaultBillingType holds the default value on creation for the billing_type field.
 	usagelog.DefaultBillingType = usagelogDescBillingType.Default.(int8)
 	// usagelogDescStream is the schema descriptor for stream field.
-	usagelogDescStream := usagelogFields[22].Descriptor()
+	usagelogDescStream := usagelogFields[23].Descriptor()
 	// usagelog.DefaultStream holds the default value on creation for the stream field.
 	usagelog.DefaultStream = usagelogDescStream.Default.(bool)
 	// usagelogDescUserAgent is the schema descriptor for user_agent field.
-	usagelogDescUserAgent := usagelogFields[25].Descriptor()
+	usagelogDescUserAgent := usagelogFields[26].Descriptor()
 	// usagelog.UserAgentValidator is a validator for the "user_agent" field. It is called by the builders before save.
 	usagelog.UserAgentValidator = usagelogDescUserAgent.Validators[0].(func(string) error)
 	// usagelogDescIPAddress is the schema descriptor for ip_address field.
-	usagelogDescIPAddress := usagelogFields[26].Descriptor()
+	usagelogDescIPAddress := usagelogFields[27].Descriptor()
 	// usagelog.IPAddressValidator is a validator for the "ip_address" field. It is called by the builders before save.
 	usagelog.IPAddressValidator = usagelogDescIPAddress.Validators[0].(func(string) error)
 	// usagelogDescImageCount is the schema descriptor for image_count field.
-	usagelogDescImageCount := usagelogFields[27].Descriptor()
+	usagelogDescImageCount := usagelogFields[28].Descriptor()
 	// usagelog.DefaultImageCount holds the default value on creation for the image_count field.
 	usagelog.DefaultImageCount = usagelogDescImageCount.Default.(int)
 	// usagelogDescImageSize is the schema descriptor for image_size field.
-	usagelogDescImageSize := usagelogFields[28].Descriptor()
+	usagelogDescImageSize := usagelogFields[29].Descriptor()
 	// usagelog.ImageSizeValidator is a validator for the "image_size" field. It is called by the builders before save.
 	usagelog.ImageSizeValidator = usagelogDescImageSize.Validators[0].(func(string) error)
 	// usagelogDescMediaType is the schema descriptor for media_type field.
-	usagelogDescMediaType := usagelogFields[29].Descriptor()
+	usagelogDescMediaType := usagelogFields[30].Descriptor()
 	// usagelog.MediaTypeValidator is a validator for the "media_type" field. It is called by the builders before save.
 	usagelog.MediaTypeValidator = usagelogDescMediaType.Validators[0].(func(string) error)
 	// usagelogDescCacheTTLOverridden is the schema descriptor for cache_ttl_overridden field.
-	usagelogDescCacheTTLOverridden := usagelogFields[30].Descriptor()
+	usagelogDescCacheTTLOverridden := usagelogFields[31].Descriptor()
 	// usagelog.DefaultCacheTTLOverridden holds the default value on creation for the cache_ttl_overridden field.
 	usagelog.DefaultCacheTTLOverridden = usagelogDescCacheTTLOverridden.Default.(bool)
 	// usagelogDescCreatedAt is the schema descriptor for created_at field.
-	usagelogDescCreatedAt := usagelogFields[31].Descriptor()
+	usagelogDescCreatedAt := usagelogFields[32].Descriptor()
 	// usagelog.DefaultCreatedAt holds the default value on creation for the created_at field.
 	usagelog.DefaultCreatedAt = usagelogDescCreatedAt.Default.(func() time.Time)
 	userMixin := schema.User{}.Mixin()

backend/ent/schema/usage_log.go

@@ -41,6 +41,12 @@ func (UsageLog) Fields() []ent.Field {
 		field.String("model").
 			MaxLen(100).
 			NotEmpty(),
+		// UpstreamModel stores the actual upstream model name when model mapping
+		// is applied. NULL means no mapping — the requested model was used as-is.
+		field.String("upstream_model").
+			MaxLen(100).
+			Optional().
+			Nillable(),
 		field.Int64("group_id").
 			Optional().
 			Nillable(),

backend/ent/usagelog.go

@@ -32,6 +32,8 @@ type UsageLog struct {
 	RequestID string `json:"request_id,omitempty"`
 	// Model holds the value of the "model" field.
 	Model string `json:"model,omitempty"`
+	// UpstreamModel holds the value of the "upstream_model" field.
+	UpstreamModel *string `json:"upstream_model,omitempty"`
 	// GroupID holds the value of the "group_id" field.
 	GroupID *int64 `json:"group_id,omitempty"`
 	// SubscriptionID holds the value of the "subscription_id" field.
@@ -175,7 +177,7 @@ func (*UsageLog) scanValues(columns []string) ([]any, error) {
 			values[i] = new(sql.NullFloat64)
 		case usagelog.FieldID, usagelog.FieldUserID, usagelog.FieldAPIKeyID, usagelog.FieldAccountID, usagelog.FieldGroupID, usagelog.FieldSubscriptionID, usagelog.FieldInputTokens, usagelog.FieldOutputTokens, usagelog.FieldCacheCreationTokens, usagelog.FieldCacheReadTokens, usagelog.FieldCacheCreation5mTokens, usagelog.FieldCacheCreation1hTokens, usagelog.FieldBillingType, usagelog.FieldDurationMs, usagelog.FieldFirstTokenMs, usagelog.FieldImageCount:
 			values[i] = new(sql.NullInt64)
-		case usagelog.FieldRequestID, usagelog.FieldModel, usagelog.FieldUserAgent, usagelog.FieldIPAddress, usagelog.FieldImageSize, usagelog.FieldMediaType:
+		case usagelog.FieldRequestID, usagelog.FieldModel, usagelog.FieldUpstreamModel, usagelog.FieldUserAgent, usagelog.FieldIPAddress, usagelog.FieldImageSize, usagelog.FieldMediaType:
 			values[i] = new(sql.NullString)
 		case usagelog.FieldCreatedAt:
 			values[i] = new(sql.NullTime)
@@ -230,6 +232,13 @@ func (_m *UsageLog) assignValues(columns []string, values []any) error {
 			} else if value.Valid {
 				_m.Model = value.String
 			}
+		case usagelog.FieldUpstreamModel:
+			if value, ok := values[i].(*sql.NullString); !ok {
+				return fmt.Errorf("unexpected type %T for field upstream_model", values[i])
+			} else if value.Valid {
+				_m.UpstreamModel = new(string)
+				*_m.UpstreamModel = value.String
+			}
 		case usagelog.FieldGroupID:
 			if value, ok := values[i].(*sql.NullInt64); !ok {
 				return fmt.Errorf("unexpected type %T for field group_id", values[i])
@@ -477,6 +486,11 @@ func (_m *UsageLog) String() string {
 	builder.WriteString("model=")
 	builder.WriteString(_m.Model)
 	builder.WriteString(", ")
+	if v := _m.UpstreamModel; v != nil {
+		builder.WriteString("upstream_model=")
+		builder.WriteString(*v)
+	}
+	builder.WriteString(", ")
 	if v := _m.GroupID; v != nil {
 		builder.WriteString("group_id=")
 		builder.WriteString(fmt.Sprintf("%v", *v))

backend/ent/usagelog/usagelog.go

@@ -24,6 +24,8 @@ const (
 	FieldRequestID = "request_id"
 	// FieldModel holds the string denoting the model field in the database.
 	FieldModel = "model"
+	// FieldUpstreamModel holds the string denoting the upstream_model field in the database.
+	FieldUpstreamModel = "upstream_model"
 	// FieldGroupID holds the string denoting the group_id field in the database.
 	FieldGroupID = "group_id"
 	// FieldSubscriptionID holds the string denoting the subscription_id field in the database.
@@ -135,6 +137,7 @@ var Columns = []string{
 	FieldAccountID,
 	FieldRequestID,
 	FieldModel,
+	FieldUpstreamModel,
 	FieldGroupID,
 	FieldSubscriptionID,
 	FieldInputTokens,
@@ -179,6 +182,8 @@ var (
 	RequestIDValidator func(string) error
 	// ModelValidator is a validator for the "model" field. It is called by the builders before save.
 	ModelValidator func(string) error
+	// UpstreamModelValidator is a validator for the "upstream_model" field. It is called by the builders before save.
+	UpstreamModelValidator func(string) error
 	// DefaultInputTokens holds the default value on creation for the "input_tokens" field.
 	DefaultInputTokens int
 	// DefaultOutputTokens holds the default value on creation for the "output_tokens" field.
@@ -258,6 +263,11 @@ func ByModel(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldModel, opts...).ToFunc()
 }
 
+// ByUpstreamModel orders the results by the upstream_model field.
+func ByUpstreamModel(opts ...sql.OrderTermOption) OrderOption {
+	return sql.OrderByField(FieldUpstreamModel, opts...).ToFunc()
+}
+
 // ByGroupID orders the results by the group_id field.
 func ByGroupID(opts ...sql.OrderTermOption) OrderOption {
 	return sql.OrderByField(FieldGroupID, opts...).ToFunc()

backend/ent/usagelog/where.go

@@ -80,6 +80,11 @@ func Model(v string) predicate.UsageLog {
 	return predicate.UsageLog(sql.FieldEQ(FieldModel, v))
 }
 
+// UpstreamModel applies equality check predicate on the "upstream_model" field. It's identical to UpstreamModelEQ.
+func UpstreamModel(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldEQ(FieldUpstreamModel, v))
+}
+
 // GroupID applies equality check predicate on the "group_id" field. It's identical to GroupIDEQ.
 func GroupID(v int64) predicate.UsageLog {
 	return predicate.UsageLog(sql.FieldEQ(FieldGroupID, v))
@@ -405,6 +410,81 @@ func ModelContainsFold(v string) predicate.UsageLog {
 	return predicate.UsageLog(sql.FieldContainsFold(FieldModel, v))
 }
 
+// UpstreamModelEQ applies the EQ predicate on the "upstream_model" field.
+func UpstreamModelEQ(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldEQ(FieldUpstreamModel, v))
+}
+
+// UpstreamModelNEQ applies the NEQ predicate on the "upstream_model" field.
+func UpstreamModelNEQ(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldNEQ(FieldUpstreamModel, v))
+}
+
+// UpstreamModelIn applies the In predicate on the "upstream_model" field.
+func UpstreamModelIn(vs ...string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldIn(FieldUpstreamModel, vs...))
+}
+
+// UpstreamModelNotIn applies the NotIn predicate on the "upstream_model" field.
+func UpstreamModelNotIn(vs ...string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldNotIn(FieldUpstreamModel, vs...))
+}
+
+// UpstreamModelGT applies the GT predicate on the "upstream_model" field.
+func UpstreamModelGT(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldGT(FieldUpstreamModel, v))
+}
+
+// UpstreamModelGTE applies the GTE predicate on the "upstream_model" field.
+func UpstreamModelGTE(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldGTE(FieldUpstreamModel, v))
+}
+
+// UpstreamModelLT applies the LT predicate on the "upstream_model" field.
+func UpstreamModelLT(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldLT(FieldUpstreamModel, v))
+}
+
+// UpstreamModelLTE applies the LTE predicate on the "upstream_model" field.
+func UpstreamModelLTE(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldLTE(FieldUpstreamModel, v))
+}
+
+// UpstreamModelContains applies the Contains predicate on the "upstream_model" field.
+func UpstreamModelContains(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldContains(FieldUpstreamModel, v))
+}
+
+// UpstreamModelHasPrefix applies the HasPrefix predicate on the "upstream_model" field.
+func UpstreamModelHasPrefix(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldHasPrefix(FieldUpstreamModel, v))
+}
+
+// UpstreamModelHasSuffix applies the HasSuffix predicate on the "upstream_model" field.
+func UpstreamModelHasSuffix(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldHasSuffix(FieldUpstreamModel, v))
+}
+
+// UpstreamModelIsNil applies the IsNil predicate on the "upstream_model" field.
+func UpstreamModelIsNil() predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldIsNull(FieldUpstreamModel))
+}
+
+// UpstreamModelNotNil applies the NotNil predicate on the "upstream_model" field.
+func UpstreamModelNotNil() predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldNotNull(FieldUpstreamModel))
+}
+
+// UpstreamModelEqualFold applies the EqualFold predicate on the "upstream_model" field.
+func UpstreamModelEqualFold(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldEqualFold(FieldUpstreamModel, v))
+}
+
+// UpstreamModelContainsFold applies the ContainsFold predicate on the "upstream_model" field.
+func UpstreamModelContainsFold(v string) predicate.UsageLog {
+	return predicate.UsageLog(sql.FieldContainsFold(FieldUpstreamModel, v))
+}
+
 // GroupIDEQ applies the EQ predicate on the "group_id" field.
 func GroupIDEQ(v int64) predicate.UsageLog {
 	return predicate.UsageLog(sql.FieldEQ(FieldGroupID, v))

backend/ent/usagelog_create.go

@@ -57,6 +57,20 @@ func (_c *UsageLogCreate) SetModel(v string) *UsageLogCreate {
 	return _c
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (_c *UsageLogCreate) SetUpstreamModel(v string) *UsageLogCreate {
+	_c.mutation.SetUpstreamModel(v)
+	return _c
+}
+
+// SetNillableUpstreamModel sets the "upstream_model" field if the given value is not nil.
+func (_c *UsageLogCreate) SetNillableUpstreamModel(v *string) *UsageLogCreate {
+	if v != nil {
+		_c.SetUpstreamModel(*v)
+	}
+	return _c
+}
+
 // SetGroupID sets the "group_id" field.
 func (_c *UsageLogCreate) SetGroupID(v int64) *UsageLogCreate {
 	_c.mutation.SetGroupID(v)
@@ -596,6 +610,11 @@ func (_c *UsageLogCreate) check() error {
 			return &ValidationError{Name: "model", err: fmt.Errorf(`ent: validator failed for field "UsageLog.model": %w`, err)}
 		}
 	}
+	if v, ok := _c.mutation.UpstreamModel(); ok {
+		if err := usagelog.UpstreamModelValidator(v); err != nil {
+			return &ValidationError{Name: "upstream_model", err: fmt.Errorf(`ent: validator failed for field "UsageLog.upstream_model": %w`, err)}
+		}
+	}
 	if _, ok := _c.mutation.InputTokens(); !ok {
 		return &ValidationError{Name: "input_tokens", err: errors.New(`ent: missing required field "UsageLog.input_tokens"`)}
 	}
@@ -714,6 +733,10 @@ func (_c *UsageLogCreate) createSpec() (*UsageLog, *sqlgraph.CreateSpec) {
 		_spec.SetField(usagelog.FieldModel, field.TypeString, value)
 		_node.Model = value
 	}
+	if value, ok := _c.mutation.UpstreamModel(); ok {
+		_spec.SetField(usagelog.FieldUpstreamModel, field.TypeString, value)
+		_node.UpstreamModel = &value
+	}
 	if value, ok := _c.mutation.InputTokens(); ok {
 		_spec.SetField(usagelog.FieldInputTokens, field.TypeInt, value)
 		_node.InputTokens = value
@@ -1011,6 +1034,24 @@ func (u *UsageLogUpsert) UpdateModel() *UsageLogUpsert {
 	return u
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (u *UsageLogUpsert) SetUpstreamModel(v string) *UsageLogUpsert {
+	u.Set(usagelog.FieldUpstreamModel, v)
+	return u
+}
+
+// UpdateUpstreamModel sets the "upstream_model" field to the value that was provided on create.
+func (u *UsageLogUpsert) UpdateUpstreamModel() *UsageLogUpsert {
+	u.SetExcluded(usagelog.FieldUpstreamModel)
+	return u
+}
+
+// ClearUpstreamModel clears the value of the "upstream_model" field.
+func (u *UsageLogUpsert) ClearUpstreamModel() *UsageLogUpsert {
+	u.SetNull(usagelog.FieldUpstreamModel)
+	return u
+}
+
 // SetGroupID sets the "group_id" field.
 func (u *UsageLogUpsert) SetGroupID(v int64) *UsageLogUpsert {
 	u.Set(usagelog.FieldGroupID, v)
@@ -1600,6 +1641,27 @@ func (u *UsageLogUpsertOne) UpdateModel() *UsageLogUpsertOne {
 	})
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (u *UsageLogUpsertOne) SetUpstreamModel(v string) *UsageLogUpsertOne {
+	return u.Update(func(s *UsageLogUpsert) {
+		s.SetUpstreamModel(v)
+	})
+}
+
+// UpdateUpstreamModel sets the "upstream_model" field to the value that was provided on create.
+func (u *UsageLogUpsertOne) UpdateUpstreamModel() *UsageLogUpsertOne {
+	return u.Update(func(s *UsageLogUpsert) {
+		s.UpdateUpstreamModel()
+	})
+}
+
+// ClearUpstreamModel clears the value of the "upstream_model" field.
+func (u *UsageLogUpsertOne) ClearUpstreamModel() *UsageLogUpsertOne {
+	return u.Update(func(s *UsageLogUpsert) {
+		s.ClearUpstreamModel()
+	})
+}
+
 // SetGroupID sets the "group_id" field.
 func (u *UsageLogUpsertOne) SetGroupID(v int64) *UsageLogUpsertOne {
 	return u.Update(func(s *UsageLogUpsert) {
@@ -2434,6 +2496,27 @@ func (u *UsageLogUpsertBulk) UpdateModel() *UsageLogUpsertBulk {
 	})
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (u *UsageLogUpsertBulk) SetUpstreamModel(v string) *UsageLogUpsertBulk {
+	return u.Update(func(s *UsageLogUpsert) {
+		s.SetUpstreamModel(v)
+	})
+}
+
+// UpdateUpstreamModel sets the "upstream_model" field to the value that was provided on create.
+func (u *UsageLogUpsertBulk) UpdateUpstreamModel() *UsageLogUpsertBulk {
+	return u.Update(func(s *UsageLogUpsert) {
+		s.UpdateUpstreamModel()
+	})
+}
+
+// ClearUpstreamModel clears the value of the "upstream_model" field.
+func (u *UsageLogUpsertBulk) ClearUpstreamModel() *UsageLogUpsertBulk {
+	return u.Update(func(s *UsageLogUpsert) {
+		s.ClearUpstreamModel()
+	})
+}
+
 // SetGroupID sets the "group_id" field.
 func (u *UsageLogUpsertBulk) SetGroupID(v int64) *UsageLogUpsertBulk {
 	return u.Update(func(s *UsageLogUpsert) {

backend/ent/usagelog_update.go

@@ -102,6 +102,26 @@ func (_u *UsageLogUpdate) SetNillableModel(v *string) *UsageLogUpdate {
 	return _u
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (_u *UsageLogUpdate) SetUpstreamModel(v string) *UsageLogUpdate {
+	_u.mutation.SetUpstreamModel(v)
+	return _u
+}
+
+// SetNillableUpstreamModel sets the "upstream_model" field if the given value is not nil.
+func (_u *UsageLogUpdate) SetNillableUpstreamModel(v *string) *UsageLogUpdate {
+	if v != nil {
+		_u.SetUpstreamModel(*v)
+	}
+	return _u
+}
+
+// ClearUpstreamModel clears the value of the "upstream_model" field.
+func (_u *UsageLogUpdate) ClearUpstreamModel() *UsageLogUpdate {
+	_u.mutation.ClearUpstreamModel()
+	return _u
+}
+
 // SetGroupID sets the "group_id" field.
 func (_u *UsageLogUpdate) SetGroupID(v int64) *UsageLogUpdate {
 	_u.mutation.SetGroupID(v)
@@ -745,6 +765,11 @@ func (_u *UsageLogUpdate) check() error {
 			return &ValidationError{Name: "model", err: fmt.Errorf(`ent: validator failed for field "UsageLog.model": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.UpstreamModel(); ok {
+		if err := usagelog.UpstreamModelValidator(v); err != nil {
+			return &ValidationError{Name: "upstream_model", err: fmt.Errorf(`ent: validator failed for field "UsageLog.upstream_model": %w`, err)}
+		}
+	}
 	if v, ok := _u.mutation.UserAgent(); ok {
 		if err := usagelog.UserAgentValidator(v); err != nil {
 			return &ValidationError{Name: "user_agent", err: fmt.Errorf(`ent: validator failed for field "UsageLog.user_agent": %w`, err)}
@@ -795,6 +820,12 @@ func (_u *UsageLogUpdate) sqlSave(ctx context.Context) (_node int, err error) {
 	if value, ok := _u.mutation.Model(); ok {
 		_spec.SetField(usagelog.FieldModel, field.TypeString, value)
 	}
+	if value, ok := _u.mutation.UpstreamModel(); ok {
+		_spec.SetField(usagelog.FieldUpstreamModel, field.TypeString, value)
+	}
+	if _u.mutation.UpstreamModelCleared() {
+		_spec.ClearField(usagelog.FieldUpstreamModel, field.TypeString)
+	}
 	if value, ok := _u.mutation.InputTokens(); ok {
 		_spec.SetField(usagelog.FieldInputTokens, field.TypeInt, value)
 	}
@@ -1177,6 +1208,26 @@ func (_u *UsageLogUpdateOne) SetNillableModel(v *string) *UsageLogUpdateOne {
 	return _u
 }
 
+// SetUpstreamModel sets the "upstream_model" field.
+func (_u *UsageLogUpdateOne) SetUpstreamModel(v string) *UsageLogUpdateOne {
+	_u.mutation.SetUpstreamModel(v)
+	return _u
+}
+
+// SetNillableUpstreamModel sets the "upstream_model" field if the given value is not nil.
+func (_u *UsageLogUpdateOne) SetNillableUpstreamModel(v *string) *UsageLogUpdateOne {
+	if v != nil {
+		_u.SetUpstreamModel(*v)
+	}
+	return _u
+}
+
+// ClearUpstreamModel clears the value of the "upstream_model" field.
+func (_u *UsageLogUpdateOne) ClearUpstreamModel() *UsageLogUpdateOne {
+	_u.mutation.ClearUpstreamModel()
+	return _u
+}
+
 // SetGroupID sets the "group_id" field.
 func (_u *UsageLogUpdateOne) SetGroupID(v int64) *UsageLogUpdateOne {
 	_u.mutation.SetGroupID(v)
@@ -1833,6 +1884,11 @@ func (_u *UsageLogUpdateOne) check() error {
 			return &ValidationError{Name: "model", err: fmt.Errorf(`ent: validator failed for field "UsageLog.model": %w`, err)}
 		}
 	}
+	if v, ok := _u.mutation.UpstreamModel(); ok {
+		if err := usagelog.UpstreamModelValidator(v); err != nil {
+			return &ValidationError{Name: "upstream_model", err: fmt.Errorf(`ent: validator failed for field "UsageLog.upstream_model": %w`, err)}
+		}
+	}
 	if v, ok := _u.mutation.UserAgent(); ok {
 		if err := usagelog.UserAgentValidator(v); err != nil {
 			return &ValidationError{Name: "user_agent", err: fmt.Errorf(`ent: validator failed for field "UsageLog.user_agent": %w`, err)}
@@ -1900,6 +1956,12 @@ func (_u *UsageLogUpdateOne) sqlSave(ctx context.Context) (_node *UsageLog, err
 	if value, ok := _u.mutation.Model(); ok {
 		_spec.SetField(usagelog.FieldModel, field.TypeString, value)
 	}
+	if value, ok := _u.mutation.UpstreamModel(); ok {
+		_spec.SetField(usagelog.FieldUpstreamModel, field.TypeString, value)
+	}
+	if _u.mutation.UpstreamModelCleared() {
+		_spec.ClearField(usagelog.FieldUpstreamModel, field.TypeString)
+	}
 	if value, ok := _u.mutation.InputTokens(); ok {
 		_spec.SetField(usagelog.FieldInputTokens, field.TypeInt, value)
 	}

backend/go.sum

@@ -22,8 +22,6 @@ github.com/andybalholm/brotli v1.2.0 h1:ukwgCxwYrmACq68yiUqwIWnGY0cTPox/M94sVwTo
 github.com/andybalholm/brotli v1.2.0/go.mod h1:rzTDkvFWvIrjDXZHkuS16NPggd91W3kUSvPlQ1pLaKY=
 github.com/apparentlymart/go-textseg/v15 v15.0.0 h1:uYvfpb3DyLSCGWnctWKGj857c6ew1u1fNQOlOtuGxQY=
 github.com/apparentlymart/go-textseg/v15 v15.0.0/go.mod h1:K8XmNZdhEBkdlyDdvbmmsvpAG721bKi0joRfFdHIWJ4=
-github.com/aws/aws-sdk-go-v2 v1.41.2 h1:LuT2rzqNQsauaGkPK/7813XxcZ3o3yePY0Iy891T2ls=
-github.com/aws/aws-sdk-go-v2 v1.41.2/go.mod h1:IvvlAZQXvTXznUPfRVfryiG1fbzE2NGK6m9u39YQ+S4=
 github.com/aws/aws-sdk-go-v2 v1.41.3 h1:4kQ/fa22KjDt13QCy1+bYADvdgcxpfH18f0zP542kZA=
 github.com/aws/aws-sdk-go-v2 v1.41.3/go.mod h1:mwsPRE8ceUUpiTgF7QmQIJ7lgsKUPQOUl3o72QBrE1o=
 github.com/aws/aws-sdk-go-v2/aws/protocol/eventstream v1.7.5 h1:zWFmPmgw4sveAYi1mRqG+E/g0461cJ5M4bJ8/nc6d3Q=
@@ -60,8 +58,6 @@ github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.15 h1:edCcNp9eGIUDUCrzoCu1jWA
 github.com/aws/aws-sdk-go-v2/service/ssooidc v1.35.15/go.mod h1:lyRQKED9xWfgkYC/wmmYfv7iVIM68Z5OQ88ZdcV1QbU=
 github.com/aws/aws-sdk-go-v2/service/sts v1.41.7 h1:NITQpgo9A5NrDZ57uOWj+abvXSb83BbyggcUBVksN7c=
 github.com/aws/aws-sdk-go-v2/service/sts v1.41.7/go.mod h1:sks5UWBhEuWYDPdwlnRFn1w7xWdH29Jcpe+/PJQefEs=
-github.com/aws/smithy-go v1.24.1 h1:VbyeNfmYkWoxMVpGUAbQumkODcYmfMRfZ8yQiH30SK0=
-github.com/aws/smithy-go v1.24.1/go.mod h1:LEj2LM3rBRQJxPZTB4KuzZkaZYnZPnvgIhb4pu07mx0=
 github.com/aws/smithy-go v1.24.2 h1:FzA3bu/nt/vDvmnkg+R8Xl46gmzEDam6mZ1hzmwXFng=
 github.com/aws/smithy-go v1.24.2/go.mod h1:YE2RhdIuDbA5E5bTdciG9KrW3+TiEONeUWCqxX9i1Fc=
 github.com/bdandy/go-errors v1.2.2 h1:WdFv/oukjTJCLa79UfkGmwX7ZxONAihKu4V0mLIs11Q=
@@ -98,10 +94,6 @@ github.com/cespare/xxhash/v2 v2.3.0/go.mod h1:VGX0DQ3Q6kWi7AoAeZDth3/j3BFtOZR5XL
 github.com/chenzhuoyu/base64x v0.0.0-20211019084208-fb5309c8db06/go.mod h1:DH46F32mSOjUmXrMHnKwZdA8wcEefY7UVqBKYGjpdQY=
 github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311 h1:qSGYFH7+jGhDF8vLC+iwCD4WpbV1EBDSzWkJODFLams=
 github.com/chenzhuoyu/base64x v0.0.0-20221115062448-fe3a3abad311/go.mod h1:b583jCggY9gE99b6G5LEC39OIiVsWj+R97kbl5odCEk=
-github.com/clipperhouse/stringish v0.1.1 h1:+NSqMOr3GR6k1FdRhhnXrLfztGzuG+VuFDfatpWHKCs=
-github.com/clipperhouse/stringish v0.1.1/go.mod h1:v/WhFtE1q0ovMta2+m+UbpZ+2/HEXNWYXQgCt4hdOzA=
-github.com/clipperhouse/uax29/v2 v2.5.0 h1:x7T0T4eTHDONxFJsL94uKNKPHrclyFI0lm7+w94cO8U=
-github.com/clipperhouse/uax29/v2 v2.5.0/go.mod h1:Wn1g7MK6OoeDT0vL+Q0SQLDz/KpfsVRgg6W7ihQeh4g=
 github.com/coder/websocket v1.8.14 h1:9L0p0iKiNOibykf283eHkKUHHrpG7f65OE3BhhO7v9g=
 github.com/coder/websocket v1.8.14/go.mod h1:NX3SzP+inril6yawo5CQXx8+fk145lPDC6pumgx0mVg=
 github.com/containerd/errdefs v1.0.0 h1:tg5yIfIlQIrxYtu9ajqY42W3lpS19XqdxRQeEwYG8PI=
@@ -238,8 +230,6 @@ github.com/mattn/go-colorable v0.1.13/go.mod h1:7S9/ev0klgBDR4GtXTXX8a3vIGJpMovk
 github.com/mattn/go-isatty v0.0.16/go.mod h1:kYGgaQfpe5nmfYZH+SKPsOc2e4SrIfOl2e/yFXSvRLM=
 github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
 github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
-github.com/mattn/go-runewidth v0.0.19 h1:v++JhqYnZuu5jSKrk9RbgF5v4CGUjqRfBm05byFGLdw=
-github.com/mattn/go-runewidth v0.0.19/go.mod h1:XBkDxAl56ILZc9knddidhrOlY5R/pDhgLpndooCuJAs=
 github.com/mattn/go-sqlite3 v1.14.17 h1:mCRHCLDUBXgpKAqIKsaAaAsrAlbkeomtRFKXh2L6YIM=
 github.com/mattn/go-sqlite3 v1.14.17/go.mod h1:2eHXhiwb8IkHr+BDWZGa96P6+rkvnG63S2DGjv9HUNg=
 github.com/mdelapenya/tlscert v0.2.0 h1:7H81W6Z/4weDvZBNOfQte5GpIMo0lGYEeWbkGp5LJHI=
@@ -273,8 +263,6 @@ github.com/morikuni/aec v1.0.0 h1:nP9CBfwrvYnBRgY6qfDQkygYDmYwOilePFkwzv4dU8A=
 github.com/morikuni/aec v1.0.0/go.mod h1:BbKIizmSmc5MMPqRYbxO4ZU0S0+P200+tUnFx7PXmsc=
 github.com/ncruces/go-strftime v1.0.0 h1:HMFp8mLCTPp341M/ZnA4qaf7ZlsbTc+miZjCLOFAw7w=
 github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
-github.com/olekukonko/tablewriter v0.0.5 h1:P2Ga83D34wi1o9J6Wh1mRuqd4mF/x/lgBS7N7AbDhec=
-github.com/olekukonko/tablewriter v0.0.5/go.mod h1:hPp6KlRPjbx+hW8ykQs1w3UBbZlj6HuIJcUGPhkA7kY=
 github.com/opencontainers/go-digest v1.0.0 h1:apOUWs51W5PlhuyGyz9FCeeBIOUDA/6nW8Oi/yOhh5U=
 github.com/opencontainers/go-digest v1.0.0/go.mod h1:0JzlMkj0TRzQZfJkVvzbP0HBR3IKzErnv2BNG4W4MAM=
 github.com/opencontainers/image-spec v1.1.1 h1:y0fUlFfIZhPF1W537XOLg0/fcx6zcHCJwooC2xJA040=
@@ -326,8 +314,6 @@ github.com/spf13/afero v1.11.0 h1:WJQKhtpdm3v2IzqG8VMqrr6Rf3UYpEF239Jy9wNepM8=
 github.com/spf13/afero v1.11.0/go.mod h1:GH9Y3pIexgf1MTIWtNGyogA5MwRIDXGUr+hbWNoBjkY=
 github.com/spf13/cast v1.6.0 h1:GEiTHELF+vaR5dhz3VqZfFSzZjYbgeKDpBxQVS4GYJ0=
 github.com/spf13/cast v1.6.0/go.mod h1:ancEpBxwJDODSW/UG4rDrAqiKolqNNh2DX3mk86cAdo=
-github.com/spf13/cobra v1.7.0 h1:hyqWnYt1ZQShIddO5kBpj3vu05/++x6tJ6dg8EC572I=
-github.com/spf13/cobra v1.7.0/go.mod h1:uLxZILRyS/50WlhOIKD7W6V5bgeIt+4sICxh6uRMrb0=
 github.com/spf13/pflag v1.0.5 h1:iy+VFUOCP1a+8yFto/drg2CJ5u0yRoB7fZw3DKv/JXA=
 github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An2Bg=
 github.com/spf13/viper v1.18.2 h1:LUXCnvUvSM6FXAsj6nnfc8Q2tp1dIgUfY9Kc8GsSOiQ=

backend/internal/domain/constants.go

@@ -82,8 +82,8 @@ var DefaultAntigravityModelMapping = map[string]string{
 	"claude-opus-4-5-20251101":   "claude-opus-4-6-thinking", // 迁移旧模型
 	"claude-sonnet-4-5-20250929": "claude-sonnet-4-5",
 	// Claude Haiku → Sonnet（无 Haiku 支持）
-	"claude-haiku-4-5":          "claude-sonnet-4-5",
-	"claude-haiku-4-5-20251001": "claude-sonnet-4-5",
+	"claude-haiku-4-5":          "claude-sonnet-4-6",
+	"claude-haiku-4-5-20251001": "claude-sonnet-4-6",
 	// Gemini 2.5 白名单
 	"gemini-2.5-flash":               "gemini-2.5-flash",
 	"gemini-2.5-flash-image":         "gemini-2.5-flash-image",

backend/internal/handler/admin/account_handler.go

@@ -165,6 +165,8 @@ type AccountWithConcurrency struct {
 	CurrentRPM        *int     `json:"current_rpm,omitempty"`         // 当前分钟 RPM 计数
 }
 
+const accountListGroupUngroupedQueryValue = "ungrouped"
+
 func (h *AccountHandler) buildAccountResponseWithRuntime(ctx context.Context, account *service.Account) AccountWithConcurrency {
 	item := AccountWithConcurrency{
 		Account:            dto.AccountFromService(account),
@@ -226,7 +228,20 @@ func (h *AccountHandler) List(c *gin.Context) {
 
 	var groupID int64
 	if groupIDStr := c.Query("group"); groupIDStr != "" {
-		groupID, _ = strconv.ParseInt(groupIDStr, 10, 64)
+		if groupIDStr == accountListGroupUngroupedQueryValue {
+			groupID = service.AccountListGroupUngrouped
+		} else {
+			parsedGroupID, parseErr := strconv.ParseInt(groupIDStr, 10, 64)
+			if parseErr != nil {
+				response.ErrorFrom(c, infraerrors.BadRequest("INVALID_GROUP_FILTER", "invalid group filter"))
+				return
+			}
+			if parsedGroupID < 0 {
+				response.ErrorFrom(c, infraerrors.BadRequest("INVALID_GROUP_FILTER", "invalid group filter"))
+				return
+			}
+			groupID = parsedGroupID
+		}
 	}
 
 	accounts, total, err := h.adminService.ListAccounts(c.Request.Context(), page, pageSize, platform, accountType, status, search, groupID)
@@ -1496,7 +1511,7 @@ func (h *OAuthHandler) SetupTokenCookieAuth(c *gin.Context) {
 }
 
 // GetUsage handles getting account usage information
-// GET /api/v1/admin/accounts/:id/usage
+// GET /api/v1/admin/accounts/:id/usage?source=passive|active
 func (h *AccountHandler) GetUsage(c *gin.Context) {
 	accountID, err := strconv.ParseInt(c.Param("id"), 10, 64)
 	if err != nil {
@@ -1504,7 +1519,14 @@ func (h *AccountHandler) GetUsage(c *gin.Context) {
 		return
 	}
 
-	usage, err := h.accountUsageService.GetUsage(c.Request.Context(), accountID)
+	source := c.DefaultQuery("source", "active")
+
+	var usage *service.UsageInfo
+	if source == "passive" {
+		usage, err = h.accountUsageService.GetPassiveUsage(c.Request.Context(), accountID)
+	} else {
+		usage, err = h.accountUsageService.GetUsage(c.Request.Context(), accountID)
+	}
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return

backend/internal/handler/admin/dashboard_handler.go

@@ -273,6 +273,7 @@ func (h *DashboardHandler) GetModelStats(c *gin.Context) {
 
 	// Parse optional filter params
 	var userID, apiKeyID, accountID, groupID int64
+	modelSource := usagestats.ModelSourceRequested
 	var requestType *int16
 	var stream *bool
 	var billingType *int8
@@ -297,6 +298,13 @@ func (h *DashboardHandler) GetModelStats(c *gin.Context) {
 			groupID = id
 		}
 	}
+	if rawModelSource := strings.TrimSpace(c.Query("model_source")); rawModelSource != "" {
+		if !usagestats.IsValidModelSource(rawModelSource) {
+			response.BadRequest(c, "Invalid model_source, use requested/upstream/mapping")
+			return
+		}
+		modelSource = rawModelSource
+	}
 	if requestTypeStr := strings.TrimSpace(c.Query("request_type")); requestTypeStr != "" {
 		parsed, err := service.ParseUsageRequestType(requestTypeStr)
 		if err != nil {
@@ -323,7 +331,7 @@ func (h *DashboardHandler) GetModelStats(c *gin.Context) {
 		}
 	}
 
-	stats, hit, err := h.getModelStatsCached(c.Request.Context(), startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+	stats, hit, err := h.getModelStatsCached(c.Request.Context(), startTime, endTime, userID, apiKeyID, accountID, groupID, modelSource, requestType, stream, billingType)
 	if err != nil {
 		response.Error(c, 500, "Failed to get model statistics")
 		return
@@ -619,6 +627,12 @@ func (h *DashboardHandler) GetUserBreakdown(c *gin.Context) {
 		}
 	}
 	dim.Model = c.Query("model")
+	rawModelSource := strings.TrimSpace(c.DefaultQuery("model_source", usagestats.ModelSourceRequested))
+	if !usagestats.IsValidModelSource(rawModelSource) {
+		response.BadRequest(c, "Invalid model_source, use requested/upstream/mapping")
+		return
+	}
+	dim.ModelType = rawModelSource
 	dim.Endpoint = c.Query("endpoint")
 	dim.EndpointType = c.DefaultQuery("endpoint_type", "inbound")
 

backend/internal/handler/admin/dashboard_handler_request_type_test.go

@@ -149,6 +149,28 @@ func TestDashboardModelStatsInvalidStream(t *testing.T) {
 	require.Equal(t, http.StatusBadRequest, rec.Code)
 }
 
+func TestDashboardModelStatsInvalidModelSource(t *testing.T) {
+	repo := &dashboardUsageRepoCapture{}
+	router := newDashboardRequestTypeTestRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet, "/admin/dashboard/models?model_source=invalid", nil)
+	rec := httptest.NewRecorder()
+	router.ServeHTTP(rec, req)
+
+	require.Equal(t, http.StatusBadRequest, rec.Code)
+}
+
+func TestDashboardModelStatsValidModelSource(t *testing.T) {
+	repo := &dashboardUsageRepoCapture{}
+	router := newDashboardRequestTypeTestRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet, "/admin/dashboard/models?model_source=upstream", nil)
+	rec := httptest.NewRecorder()
+	router.ServeHTTP(rec, req)
+
+	require.Equal(t, http.StatusOK, rec.Code)
+}
+
 func TestDashboardUsersRankingLimitAndCache(t *testing.T) {
 	dashboardUsersRankingCache = newSnapshotCache(5 * time.Minute)
 	repo := &dashboardUsageRepoCapture{

backend/internal/handler/admin/dashboard_handler_user_breakdown_test.go

@@ -73,9 +73,35 @@ func TestGetUserBreakdown_ModelFilter(t *testing.T) {
 
 	require.Equal(t, http.StatusOK, w.Code)
 	require.Equal(t, "claude-opus-4-6", repo.capturedDim.Model)
+	require.Equal(t, usagestats.ModelSourceRequested, repo.capturedDim.ModelType)
 	require.Equal(t, int64(0), repo.capturedDim.GroupID)
 }
 
+func TestGetUserBreakdown_ModelSourceFilter(t *testing.T) {
+	repo := &userBreakdownRepoCapture{}
+	router := newUserBreakdownRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet,
+		"/admin/dashboard/user-breakdown?start_date=2026-03-01&end_date=2026-03-16&model=claude-opus-4-6&model_source=upstream", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	require.Equal(t, http.StatusOK, w.Code)
+	require.Equal(t, usagestats.ModelSourceUpstream, repo.capturedDim.ModelType)
+}
+
+func TestGetUserBreakdown_InvalidModelSource(t *testing.T) {
+	repo := &userBreakdownRepoCapture{}
+	router := newUserBreakdownRouter(repo)
+
+	req := httptest.NewRequest(http.MethodGet,
+		"/admin/dashboard/user-breakdown?start_date=2026-03-01&end_date=2026-03-16&model_source=foobar", nil)
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	require.Equal(t, http.StatusBadRequest, w.Code)
+}
+
 func TestGetUserBreakdown_EndpointFilter(t *testing.T) {
 	repo := &userBreakdownRepoCapture{}
 	router := newUserBreakdownRouter(repo)

backend/internal/handler/admin/dashboard_query_cache.go

@@ -38,6 +38,7 @@ type dashboardModelGroupCacheKey struct {
 	APIKeyID    int64  `json:"api_key_id"`
 	AccountID   int64  `json:"account_id"`
 	GroupID     int64  `json:"group_id"`
+	ModelSource string `json:"model_source,omitempty"`
 	RequestType *int16 `json:"request_type"`
 	Stream      *bool  `json:"stream"`
 	BillingType *int8  `json:"billing_type"`
@@ -111,6 +112,7 @@ func (h *DashboardHandler) getModelStatsCached(
 	ctx context.Context,
 	startTime, endTime time.Time,
 	userID, apiKeyID, accountID, groupID int64,
+	modelSource string,
 	requestType *int16,
 	stream *bool,
 	billingType *int8,
@@ -122,12 +124,13 @@ func (h *DashboardHandler) getModelStatsCached(
 		APIKeyID:    apiKeyID,
 		AccountID:   accountID,
 		GroupID:     groupID,
+		ModelSource: usagestats.NormalizeModelSource(modelSource),
 		RequestType: requestType,
 		Stream:      stream,
 		BillingType: billingType,
 	})
 	entry, hit, err := dashboardModelStatsCache.GetOrLoad(key, func() (any, error) {
-		return h.dashboardService.GetModelStatsWithFilters(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+		return h.dashboardService.GetModelStatsWithFiltersBySource(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType, modelSource)
 	})
 	if err != nil {
 		return nil, hit, err

backend/internal/handler/admin/dashboard_snapshot_v2_handler.go

@@ -200,6 +200,7 @@ func (h *DashboardHandler) buildSnapshotV2Response(
 			filters.APIKeyID,
 			filters.AccountID,
 			filters.GroupID,
+			usagestats.ModelSourceRequested,
 			filters.RequestType,
 			filters.Stream,
 			filters.BillingType,

backend/internal/handler/admin/setting_handler.go

@@ -977,6 +977,58 @@ func (h *SettingHandler) DeleteAdminAPIKey(c *gin.Context) {
 	response.Success(c, gin.H{"message": "Admin API key deleted"})
 }
 
+// GetOverloadCooldownSettings 获取529过载冷却配置
+// GET /api/v1/admin/settings/overload-cooldown
+func (h *SettingHandler) GetOverloadCooldownSettings(c *gin.Context) {
+	settings, err := h.settingService.GetOverloadCooldownSettings(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, dto.OverloadCooldownSettings{
+		Enabled:         settings.Enabled,
+		CooldownMinutes: settings.CooldownMinutes,
+	})
+}
+
+// UpdateOverloadCooldownSettingsRequest 更新529过载冷却配置请求
+type UpdateOverloadCooldownSettingsRequest struct {
+	Enabled         bool `json:"enabled"`
+	CooldownMinutes int  `json:"cooldown_minutes"`
+}
+
+// UpdateOverloadCooldownSettings 更新529过载冷却配置
+// PUT /api/v1/admin/settings/overload-cooldown
+func (h *SettingHandler) UpdateOverloadCooldownSettings(c *gin.Context) {
+	var req UpdateOverloadCooldownSettingsRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+
+	settings := &service.OverloadCooldownSettings{
+		Enabled:         req.Enabled,
+		CooldownMinutes: req.CooldownMinutes,
+	}
+
+	if err := h.settingService.SetOverloadCooldownSettings(c.Request.Context(), settings); err != nil {
+		response.BadRequest(c, err.Error())
+		return
+	}
+
+	updatedSettings, err := h.settingService.GetOverloadCooldownSettings(c.Request.Context())
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, dto.OverloadCooldownSettings{
+		Enabled:         updatedSettings.Enabled,
+		CooldownMinutes: updatedSettings.CooldownMinutes,
+	})
+}
+
 // GetStreamTimeoutSettings 获取流超时处理配置
 // GET /api/v1/admin/settings/stream-timeout
 func (h *SettingHandler) GetStreamTimeoutSettings(c *gin.Context) {

backend/internal/handler/dto/mappers.go

@@ -523,6 +523,7 @@ func usageLogFromServiceUser(l *service.UsageLog) UsageLog {
 		AccountID:             l.AccountID,
 		RequestID:             l.RequestID,
 		Model:                 l.Model,
+		UpstreamModel:         l.UpstreamModel,
 		ServiceTier:           l.ServiceTier,
 		ReasoningEffort:       l.ReasoningEffort,
 		InboundEndpoint:       l.InboundEndpoint,

backend/internal/handler/dto/settings.go

@@ -157,6 +157,12 @@ type ListSoraS3ProfilesResponse struct {
 	Items           []SoraS3Profile `json:"items"`
 }
 
+// OverloadCooldownSettings 529过载冷却配置 DTO
+type OverloadCooldownSettings struct {
+	Enabled         bool `json:"enabled"`
+	CooldownMinutes int  `json:"cooldown_minutes"`
+}
+
 // StreamTimeoutSettings 流超时处理配置 DTO
 type StreamTimeoutSettings struct {
 	Enabled                bool   `json:"enabled"`

backend/internal/handler/dto/types.go

@@ -334,6 +334,9 @@ type UsageLog struct {
 	AccountID int64  `json:"account_id"`
 	RequestID string `json:"request_id"`
 	Model     string `json:"model"`
+	// UpstreamModel is the actual model sent to the upstream provider after mapping.
+	// Omitted when no mapping was applied (requested model was used as-is).
+	UpstreamModel *string `json:"upstream_model,omitempty"`
 	// ServiceTier records the OpenAI service tier used for billing, e.g. "priority" / "flex".
 	ServiceTier *string `json:"service_tier,omitempty"`
 	// ReasoningEffort is the request's reasoning effort level.

backend/internal/handler/gateway_handler.go

@@ -1219,6 +1219,10 @@ func (h *GatewayHandler) handleFailoverExhausted(c *gin.Context, failoverErr *se
 		}
 	}
 
+	// 记录原始上游状态码，以便 ops 错误日志捕获真实的上游错误
+	upstreamMsg := service.ExtractUpstreamErrorMessage(responseBody)
+	service.SetOpsUpstreamError(c, statusCode, upstreamMsg, "")
+
 	// 使用默认的错误映射
 	status, errType, errMsg := h.mapUpstreamError(statusCode)
 	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
@@ -1227,6 +1231,7 @@ func (h *GatewayHandler) handleFailoverExhausted(c *gin.Context, failoverErr *se
 // handleFailoverExhaustedSimple 简化版本，用于没有响应体的情况
 func (h *GatewayHandler) handleFailoverExhaustedSimple(c *gin.Context, statusCode int, streamStarted bool) {
 	status, errType, errMsg := h.mapUpstreamError(statusCode)
+	service.SetOpsUpstreamError(c, statusCode, errMsg, "")
 	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
 }
 

backend/internal/handler/gemini_v1beta_handler.go

@@ -593,6 +593,10 @@ func (h *GatewayHandler) handleGeminiFailoverExhausted(c *gin.Context, failoverE
 		}
 	}
 
+	// 记录原始上游状态码，以便 ops 错误日志捕获真实的上游错误
+	upstreamMsg := service.ExtractUpstreamErrorMessage(responseBody)
+	service.SetOpsUpstreamError(c, statusCode, upstreamMsg, "")
+
 	// 使用默认的错误映射
 	status, message := mapGeminiUpstreamError(statusCode)
 	googleError(c, status, message)

backend/internal/handler/openai_gateway_handler.go

@@ -1435,6 +1435,10 @@ func (h *OpenAIGatewayHandler) handleFailoverExhausted(c *gin.Context, failoverE
 		}
 	}
 
+	// 记录原始上游状态码，以便 ops 错误日志捕获真实的上游错误
+	upstreamMsg := service.ExtractUpstreamErrorMessage(responseBody)
+	service.SetOpsUpstreamError(c, statusCode, upstreamMsg, "")
+
 	// 使用默认的错误映射
 	status, errType, errMsg := h.mapUpstreamError(statusCode)
 	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
@@ -1443,6 +1447,7 @@ func (h *OpenAIGatewayHandler) handleFailoverExhausted(c *gin.Context, failoverE
 // handleFailoverExhaustedSimple 简化版本，用于没有响应体的情况
 func (h *OpenAIGatewayHandler) handleFailoverExhaustedSimple(c *gin.Context, statusCode int, streamStarted bool) {
 	status, errType, errMsg := h.mapUpstreamError(statusCode)
+	service.SetOpsUpstreamError(c, statusCode, errMsg, "")
 	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
 }
 

backend/internal/handler/sora_gateway_handler.go

@@ -484,6 +484,9 @@ func (h *SoraGatewayHandler) handleConcurrencyError(c *gin.Context, err error, s
 }
 
 func (h *SoraGatewayHandler) handleFailoverExhausted(c *gin.Context, statusCode int, responseHeaders http.Header, responseBody []byte, streamStarted bool) {
+	upstreamMsg := service.ExtractUpstreamErrorMessage(responseBody)
+	service.SetOpsUpstreamError(c, statusCode, upstreamMsg, "")
+
 	status, errType, errMsg := h.mapUpstreamError(statusCode, responseHeaders, responseBody)
 	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
 }

backend/internal/pkg/antigravity/request_transformer.go

@@ -275,21 +275,6 @@ func filterOpenCodePrompt(text string) string {
 	return ""
 }
 
-// systemBlockFilterPrefixes 需要从 system 中过滤的文本前缀列表
-var systemBlockFilterPrefixes = []string{
-	"x-anthropic-billing-header",
-}
-
-// filterSystemBlockByPrefix 如果文本匹配过滤前缀，返回空字符串
-func filterSystemBlockByPrefix(text string) string {
-	for _, prefix := range systemBlockFilterPrefixes {
-		if strings.HasPrefix(text, prefix) {
-			return ""
-		}
-	}
-	return text
-}
-
 // buildSystemInstruction 构建 systemInstruction（与 Antigravity-Manager 保持一致）
 func buildSystemInstruction(system json.RawMessage, modelName string, opts TransformOptions, tools []ClaudeTool) *GeminiContent {
 	var parts []GeminiPart
@@ -306,8 +291,8 @@ func buildSystemInstruction(system json.RawMessage, modelName string, opts Trans
 				if strings.Contains(sysStr, "You are Antigravity") {
 					userHasAntigravityIdentity = true
 				}
-				// 过滤 OpenCode 默认提示词和黑名单前缀
-				filtered := filterSystemBlockByPrefix(filterOpenCodePrompt(sysStr))
+				// 过滤 OpenCode 默认提示词
+				filtered := filterOpenCodePrompt(sysStr)
 				if filtered != "" {
 					userSystemParts = append(userSystemParts, GeminiPart{Text: filtered})
 				}
@@ -321,8 +306,8 @@ func buildSystemInstruction(system json.RawMessage, modelName string, opts Trans
 						if strings.Contains(block.Text, "You are Antigravity") {
 							userHasAntigravityIdentity = true
 						}
-						// 过滤 OpenCode 默认提示词和黑名单前缀
-						filtered := filterSystemBlockByPrefix(filterOpenCodePrompt(block.Text))
+						// 过滤 OpenCode 默认提示词
+						filtered := filterOpenCodePrompt(block.Text)
 						if filtered != "" {
 							userSystemParts = append(userSystemParts, GeminiPart{Text: filtered})
 						}

backend/internal/pkg/antigravity/request_transformer_test.go

@@ -2,7 +2,10 @@ package antigravity
 
 import (
 	"encoding/json"
+	"strings"
 	"testing"
+
+	"github.com/stretchr/testify/require"
 )
 
 // TestBuildParts_ThinkingBlockWithoutSignature 测试thinking block无signature时的处理
@@ -349,3 +352,51 @@ func TestBuildGenerationConfig_ThinkingDynamicBudget(t *testing.T) {
 		})
 	}
 }
+
+func TestTransformClaudeToGeminiWithOptions_PreservesBillingHeaderSystemBlock(t *testing.T) {
+	tests := []struct {
+		name   string
+		system json.RawMessage
+	}{
+		{
+			name:   "system array",
+			system: json.RawMessage(`[{"type":"text","text":"x-anthropic-billing-header keep"}]`),
+		},
+		{
+			name:   "system string",
+			system: json.RawMessage(`"x-anthropic-billing-header keep"`),
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			claudeReq := &ClaudeRequest{
+				Model:  "claude-3-5-sonnet-latest",
+				System: tt.system,
+				Messages: []ClaudeMessage{
+					{
+						Role:    "user",
+						Content: json.RawMessage(`[{"type":"text","text":"hello"}]`),
+					},
+				},
+			}
+
+			body, err := TransformClaudeToGeminiWithOptions(claudeReq, "project-1", "gemini-2.5-flash", DefaultTransformOptions())
+			require.NoError(t, err)
+
+			var req V1InternalRequest
+			require.NoError(t, json.Unmarshal(body, &req))
+			require.NotNil(t, req.Request.SystemInstruction)
+
+			found := false
+			for _, part := range req.Request.SystemInstruction.Parts {
+				if strings.Contains(part.Text, "x-anthropic-billing-header keep") {
+					found = true
+					break
+				}
+			}
+
+			require.True(t, found, "转换后的 systemInstruction 应保留 x-anthropic-billing-header 内容")
+		})
+	}
+}

backend/internal/pkg/apicompat/anthropic_responses_test.go

@@ -1008,3 +1008,114 @@ func TestAnthropicToResponses_ImageEmptyMediaType(t *testing.T) {
 	// Should default to image/png when media_type is empty.
 	assert.Equal(t, "data:image/png;base64,iVBOR", parts[0].ImageURL)
 }
+
+// ---------------------------------------------------------------------------
+// normalizeToolParameters tests
+// ---------------------------------------------------------------------------
+
+func TestNormalizeToolParameters(t *testing.T) {
+	tests := []struct {
+		name     string
+		input    json.RawMessage
+		expected string
+	}{
+		{
+			name:     "nil input",
+			input:    nil,
+			expected: `{"type":"object","properties":{}}`,
+		},
+		{
+			name:     "empty input",
+			input:    json.RawMessage(``),
+			expected: `{"type":"object","properties":{}}`,
+		},
+		{
+			name:     "null input",
+			input:    json.RawMessage(`null`),
+			expected: `{"type":"object","properties":{}}`,
+		},
+		{
+			name:     "object without properties",
+			input:    json.RawMessage(`{"type":"object"}`),
+			expected: `{"type":"object","properties":{}}`,
+		},
+		{
+			name:     "object with properties",
+			input:    json.RawMessage(`{"type":"object","properties":{"city":{"type":"string"}}}`),
+			expected: `{"type":"object","properties":{"city":{"type":"string"}}}`,
+		},
+		{
+			name:     "non-object type",
+			input:    json.RawMessage(`{"type":"string"}`),
+			expected: `{"type":"string"}`,
+		},
+		{
+			name:     "object with additional fields preserved",
+			input:    json.RawMessage(`{"type":"object","required":["name"]}`),
+			expected: `{"type":"object","required":["name"],"properties":{}}`,
+		},
+		{
+			name:     "invalid JSON passthrough",
+			input:    json.RawMessage(`not json`),
+			expected: `not json`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			result := normalizeToolParameters(tt.input)
+			if tt.name == "invalid JSON passthrough" {
+				assert.Equal(t, tt.expected, string(result))
+			} else {
+				assert.JSONEq(t, tt.expected, string(result))
+			}
+		})
+	}
+}
+
+func TestAnthropicToResponses_ToolWithoutProperties(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages: []AnthropicMessage{
+			{Role: "user", Content: json.RawMessage(`"Hello"`)},
+		},
+		Tools: []AnthropicTool{
+			{Name: "mcp__pencil__get_style_guide_tags", Description: "Get style tags", InputSchema: json.RawMessage(`{"type":"object"}`)},
+		},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	require.Len(t, resp.Tools, 1)
+	assert.Equal(t, "function", resp.Tools[0].Type)
+	assert.Equal(t, "mcp__pencil__get_style_guide_tags", resp.Tools[0].Name)
+
+	// Parameters must have "properties" field after normalization.
+	var params map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(resp.Tools[0].Parameters, &params))
+	assert.Contains(t, params, "properties")
+}
+
+func TestAnthropicToResponses_ToolWithNilSchema(t *testing.T) {
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages: []AnthropicMessage{
+			{Role: "user", Content: json.RawMessage(`"Hello"`)},
+		},
+		Tools: []AnthropicTool{
+			{Name: "simple_tool", Description: "A tool"},
+		},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+
+	require.Len(t, resp.Tools, 1)
+	var params map[string]json.RawMessage
+	require.NoError(t, json.Unmarshal(resp.Tools[0].Parameters, &params))
+	assert.JSONEq(t, `"object"`, string(params["type"]))
+	assert.JSONEq(t, `{}`, string(params["properties"]))
+}

backend/internal/pkg/apicompat/anthropic_to_responses.go

@@ -409,8 +409,41 @@ func convertAnthropicToolsToResponses(tools []AnthropicTool) []ResponsesTool {
 			Type:        "function",
 			Name:        t.Name,
 			Description: t.Description,
-			Parameters:  t.InputSchema,
+			Parameters:  normalizeToolParameters(t.InputSchema),
 		})
 	}
 	return out
 }
+
+// normalizeToolParameters ensures the tool parameter schema is valid for
+// OpenAI's Responses API, which requires "properties" on object schemas.
+//
+//   - nil/empty → {"type":"object","properties":{}}
+//   - type=object without properties → adds "properties": {}
+//   - otherwise → returned unchanged
+func normalizeToolParameters(schema json.RawMessage) json.RawMessage {
+	if len(schema) == 0 || string(schema) == "null" {
+		return json.RawMessage(`{"type":"object","properties":{}}`)
+	}
+
+	var m map[string]json.RawMessage
+	if err := json.Unmarshal(schema, &m); err != nil {
+		return schema
+	}
+
+	typ := m["type"]
+	if string(typ) != `"object"` {
+		return schema
+	}
+
+	if _, ok := m["properties"]; ok {
+		return schema
+	}
+
+	m["properties"] = json.RawMessage(`{}`)
+	out, err := json.Marshal(m)
+	if err != nil {
+		return schema
+	}
+	return out
+}

backend/internal/pkg/usagestats/usage_log_types.go

@@ -3,6 +3,28 @@ package usagestats
 
 import "time"
 
+const (
+	ModelSourceRequested = "requested"
+	ModelSourceUpstream  = "upstream"
+	ModelSourceMapping   = "mapping"
+)
+
+func IsValidModelSource(source string) bool {
+	switch source {
+	case ModelSourceRequested, ModelSourceUpstream, ModelSourceMapping:
+		return true
+	default:
+		return false
+	}
+}
+
+func NormalizeModelSource(source string) string {
+	if IsValidModelSource(source) {
+		return source
+	}
+	return ModelSourceRequested
+}
+
 // DashboardStats 仪表盘统计
 type DashboardStats struct {
 	// 用户统计
@@ -150,6 +172,7 @@ type UserBreakdownItem struct {
 type UserBreakdownDimension struct {
 	GroupID      int64  // filter by group_id (>0 to enable)
 	Model        string // filter by model name (non-empty to enable)
+	ModelType    string // "requested", "upstream", or "mapping"
 	Endpoint     string // filter by endpoint value (non-empty to enable)
 	EndpointType string // "inbound", "upstream", or "path"
 }

backend/internal/pkg/usagestats/usage_log_types_test.go

@@ -0,0 +1,47 @@
+package usagestats
+
+import "testing"
+
+func TestIsValidModelSource(t *testing.T) {
+	tests := []struct {
+		name   string
+		source string
+		want   bool
+	}{
+		{name: "requested", source: ModelSourceRequested, want: true},
+		{name: "upstream", source: ModelSourceUpstream, want: true},
+		{name: "mapping", source: ModelSourceMapping, want: true},
+		{name: "invalid", source: "foobar", want: false},
+		{name: "empty", source: "", want: false},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := IsValidModelSource(tc.source); got != tc.want {
+				t.Fatalf("IsValidModelSource(%q)=%v want %v", tc.source, got, tc.want)
+			}
+		})
+	}
+}
+
+func TestNormalizeModelSource(t *testing.T) {
+	tests := []struct {
+		name   string
+		source string
+		want   string
+	}{
+		{name: "requested", source: ModelSourceRequested, want: ModelSourceRequested},
+		{name: "upstream", source: ModelSourceUpstream, want: ModelSourceUpstream},
+		{name: "mapping", source: ModelSourceMapping, want: ModelSourceMapping},
+		{name: "invalid falls back", source: "foobar", want: ModelSourceRequested},
+		{name: "empty falls back", source: "", want: ModelSourceRequested},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := NormalizeModelSource(tc.source); got != tc.want {
+				t.Fatalf("NormalizeModelSource(%q)=%q want %q", tc.source, got, tc.want)
+			}
+		})
+	}
+}

backend/internal/repository/account_repo.go

@@ -56,6 +56,7 @@ var schedulerNeutralExtraKeyPrefixes = []string{
 	"codex_secondary_",
 	"codex_5h_",
 	"codex_7d_",
+	"passive_usage_",
 }
 
 var schedulerNeutralExtraKeys = map[string]struct{}{
@@ -473,7 +474,9 @@ func (r *accountRepository) ListWithFilters(ctx context.Context, params paginati
 	if search != "" {
 		q = q.Where(dbaccount.NameContainsFold(search))
 	}
-	if groupID > 0 {
+	if groupID == service.AccountListGroupUngrouped {
+		q = q.Where(dbaccount.Not(dbaccount.HasAccountGroups()))
+	} else if groupID > 0 {
 		q = q.Where(dbaccount.HasAccountGroupsWith(dbaccountgroup.GroupIDEQ(groupID)))
 	}
 

backend/internal/repository/account_repo_integration_test.go

@@ -214,6 +214,7 @@ func (s *AccountRepoSuite) TestListWithFilters() {
 		accType   string
 		status    string
 		search    string
+		groupID   int64
 		wantCount int
 		validate  func(accounts []service.Account)
 	}{
@@ -265,6 +266,21 @@ func (s *AccountRepoSuite) TestListWithFilters() {
 				s.Require().Contains(accounts[0].Name, "alpha")
 			},
 		},
+		{
+			name: "filter_by_ungrouped",
+			setup: func(client *dbent.Client) {
+				group := mustCreateGroup(s.T(), client, &service.Group{Name: "g-ungrouped"})
+				grouped := mustCreateAccount(s.T(), client, &service.Account{Name: "grouped-account"})
+				mustCreateAccount(s.T(), client, &service.Account{Name: "ungrouped-account"})
+				mustBindAccountToGroup(s.T(), client, grouped.ID, group.ID, 1)
+			},
+			groupID:   service.AccountListGroupUngrouped,
+			wantCount: 1,
+			validate: func(accounts []service.Account) {
+				s.Require().Equal("ungrouped-account", accounts[0].Name)
+				s.Require().Empty(accounts[0].GroupIDs)
+			},
+		},
 	}
 
 	for _, tt := range tests {
@@ -277,7 +293,7 @@ func (s *AccountRepoSuite) TestListWithFilters() {
 
 			tt.setup(client)
 
-			accounts, _, err := repo.ListWithFilters(ctx, pagination.PaginationParams{Page: 1, PageSize: 10}, tt.platform, tt.accType, tt.status, tt.search, 0)
+			accounts, _, err := repo.ListWithFilters(ctx, pagination.PaginationParams{Page: 1, PageSize: 10}, tt.platform, tt.accType, tt.status, tt.search, tt.groupID)
 			s.Require().NoError(err)
 			s.Require().Len(accounts, tt.wantCount)
 			if tt.validate != nil {

backend/internal/repository/usage_log_repo.go

@@ -28,7 +28,7 @@ import (
 	gocache "github.com/patrickmn/go-cache"
 )
 
-const usageLogSelectColumns = "id, user_id, api_key_id, account_id, request_id, model, group_id, subscription_id, input_tokens, output_tokens, cache_creation_tokens, cache_read_tokens, cache_creation_5m_tokens, cache_creation_1h_tokens, input_cost, output_cost, cache_creation_cost, cache_read_cost, total_cost, actual_cost, rate_multiplier, account_rate_multiplier, billing_type, request_type, stream, openai_ws_mode, duration_ms, first_token_ms, user_agent, ip_address, image_count, image_size, media_type, service_tier, reasoning_effort, inbound_endpoint, upstream_endpoint, cache_ttl_overridden, created_at"
+const usageLogSelectColumns = "id, user_id, api_key_id, account_id, request_id, model, upstream_model, group_id, subscription_id, input_tokens, output_tokens, cache_creation_tokens, cache_read_tokens, cache_creation_5m_tokens, cache_creation_1h_tokens, input_cost, output_cost, cache_creation_cost, cache_read_cost, total_cost, actual_cost, rate_multiplier, account_rate_multiplier, billing_type, request_type, stream, openai_ws_mode, duration_ms, first_token_ms, user_agent, ip_address, image_count, image_size, media_type, service_tier, reasoning_effort, inbound_endpoint, upstream_endpoint, cache_ttl_overridden, created_at"
 
 var usageLogInsertArgTypes = [...]string{
 	"bigint",
@@ -36,6 +36,7 @@ var usageLogInsertArgTypes = [...]string{
 	"bigint",
 	"text",
 	"text",
+	"text",
 	"bigint",
 	"bigint",
 	"integer",
@@ -277,6 +278,7 @@ func (r *usageLogRepository) createSingle(ctx context.Context, sqlq sqlExecutor,
 			account_id,
 			request_id,
 			model,
+			upstream_model,
 			group_id,
 			subscription_id,
 			input_tokens,
@@ -311,12 +313,12 @@ func (r *usageLogRepository) createSingle(ctx context.Context, sqlq sqlExecutor,
 			cache_ttl_overridden,
 			created_at
 		) VALUES (
-			$1, $2, $3, $4, $5,
-			$6, $7,
-			$8, $9, $10, $11,
-			$12, $13,
-			$14, $15, $16, $17, $18, $19,
-			$20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33, $34, $35, $36, $37, $38
+			$1, $2, $3, $4, $5, $6,
+			$7, $8,
+			$9, $10, $11, $12,
+			$13, $14,
+			$15, $16, $17, $18, $19, $20,
+			$21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33, $34, $35, $36, $37, $38, $39
 		)
 		ON CONFLICT (request_id, api_key_id) DO NOTHING
 		RETURNING id, created_at
@@ -707,6 +709,7 @@ func buildUsageLogBatchInsertQuery(keys []string, preparedByKey map[string]usage
 			account_id,
 			request_id,
 			model,
+			upstream_model,
 			group_id,
 			subscription_id,
 			input_tokens,
@@ -742,7 +745,7 @@ func buildUsageLogBatchInsertQuery(keys []string, preparedByKey map[string]usage
 			created_at
 		) AS (VALUES `)
 
-	args := make([]any, 0, len(keys)*38)
+	args := make([]any, 0, len(keys)*39)
 	argPos := 1
 	for idx, key := range keys {
 		if idx > 0 {
@@ -776,6 +779,7 @@ func buildUsageLogBatchInsertQuery(keys []string, preparedByKey map[string]usage
 				account_id,
 				request_id,
 				model,
+				upstream_model,
 				group_id,
 				subscription_id,
 				input_tokens,
@@ -816,6 +820,7 @@ func buildUsageLogBatchInsertQuery(keys []string, preparedByKey map[string]usage
 				account_id,
 				request_id,
 				model,
+				upstream_model,
 				group_id,
 				subscription_id,
 				input_tokens,
@@ -896,6 +901,7 @@ func buildUsageLogBestEffortInsertQuery(preparedList []usageLogInsertPrepared) (
 			account_id,
 			request_id,
 			model,
+			upstream_model,
 			group_id,
 			subscription_id,
 			input_tokens,
@@ -931,7 +937,7 @@ func buildUsageLogBestEffortInsertQuery(preparedList []usageLogInsertPrepared) (
 			created_at
 		) AS (VALUES `)
 
-	args := make([]any, 0, len(preparedList)*38)
+	args := make([]any, 0, len(preparedList)*39)
 	argPos := 1
 	for idx, prepared := range preparedList {
 		if idx > 0 {
@@ -962,6 +968,7 @@ func buildUsageLogBestEffortInsertQuery(preparedList []usageLogInsertPrepared) (
 			account_id,
 			request_id,
 			model,
+			upstream_model,
 			group_id,
 			subscription_id,
 			input_tokens,
@@ -1002,6 +1009,7 @@ func buildUsageLogBestEffortInsertQuery(preparedList []usageLogInsertPrepared) (
 			account_id,
 			request_id,
 			model,
+			upstream_model,
 			group_id,
 			subscription_id,
 			input_tokens,
@@ -1050,6 +1058,7 @@ func execUsageLogInsertNoResult(ctx context.Context, sqlq sqlExecutor, prepared
 			account_id,
 			request_id,
 			model,
+			upstream_model,
 			group_id,
 			subscription_id,
 			input_tokens,
@@ -1084,12 +1093,12 @@ func execUsageLogInsertNoResult(ctx context.Context, sqlq sqlExecutor, prepared
 			cache_ttl_overridden,
 			created_at
 		) VALUES (
-			$1, $2, $3, $4, $5,
-			$6, $7,
-			$8, $9, $10, $11,
-			$12, $13,
-			$14, $15, $16, $17, $18, $19,
-			$20, $21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33, $34, $35, $36, $37, $38
+			$1, $2, $3, $4, $5, $6,
+			$7, $8,
+			$9, $10, $11, $12,
+			$13, $14,
+			$15, $16, $17, $18, $19, $20,
+			$21, $22, $23, $24, $25, $26, $27, $28, $29, $30, $31, $32, $33, $34, $35, $36, $37, $38, $39
 		)
 		ON CONFLICT (request_id, api_key_id) DO NOTHING
 	`, prepared.args...)
@@ -1121,6 +1130,7 @@ func prepareUsageLogInsert(log *service.UsageLog) usageLogInsertPrepared {
 	reasoningEffort := nullString(log.ReasoningEffort)
 	inboundEndpoint := nullString(log.InboundEndpoint)
 	upstreamEndpoint := nullString(log.UpstreamEndpoint)
+	upstreamModel := nullString(log.UpstreamModel)
 
 	var requestIDArg any
 	if requestID != "" {
@@ -1138,6 +1148,7 @@ func prepareUsageLogInsert(log *service.UsageLog) usageLogInsertPrepared {
 			log.AccountID,
 			requestIDArg,
 			log.Model,
+			upstreamModel,
 			groupID,
 			subscriptionID,
 			log.InputTokens,
@@ -2864,15 +2875,26 @@ func (r *usageLogRepository) getUsageTrendFromAggregates(ctx context.Context, st
 
 // GetModelStatsWithFilters returns model statistics with optional filters
 func (r *usageLogRepository) GetModelStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8) (results []ModelStat, err error) {
+	return r.getModelStatsWithFiltersBySource(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType, usagestats.ModelSourceRequested)
+}
+
+// GetModelStatsWithFiltersBySource returns model statistics with optional filters and model source dimension.
+// source: requested | upstream | mapping.
+func (r *usageLogRepository) GetModelStatsWithFiltersBySource(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8, source string) (results []ModelStat, err error) {
+	return r.getModelStatsWithFiltersBySource(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType, source)
+}
+
+func (r *usageLogRepository) getModelStatsWithFiltersBySource(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8, source string) (results []ModelStat, err error) {
 	actualCostExpr := "COALESCE(SUM(actual_cost), 0) as actual_cost"
 	// 当仅按 account_id 聚合时，实际费用使用账号倍率（total_cost * account_rate_multiplier）。
 	if accountID > 0 && userID == 0 && apiKeyID == 0 {
 		actualCostExpr = "COALESCE(SUM(total_cost * COALESCE(account_rate_multiplier, 1)), 0) as actual_cost"
 	}
+	modelExpr := resolveModelDimensionExpression(source)
 
 	query := fmt.Sprintf(`
 		SELECT
-			model,
+			%s as model,
 			COUNT(*) as requests,
 			COALESCE(SUM(input_tokens), 0) as input_tokens,
 			COALESCE(SUM(output_tokens), 0) as output_tokens,
@@ -2883,7 +2905,7 @@ func (r *usageLogRepository) GetModelStatsWithFilters(ctx context.Context, start
 			%s
 		FROM usage_logs
 		WHERE created_at >= $1 AND created_at < $2
-	`, actualCostExpr)
+	`, modelExpr, actualCostExpr)
 
 	args := []any{startTime, endTime}
 	if userID > 0 {
@@ -2907,7 +2929,7 @@ func (r *usageLogRepository) GetModelStatsWithFilters(ctx context.Context, start
 		query += fmt.Sprintf(" AND billing_type = $%d", len(args)+1)
 		args = append(args, int16(*billingType))
 	}
-	query += " GROUP BY model ORDER BY total_tokens DESC"
+	query += fmt.Sprintf(" GROUP BY %s ORDER BY total_tokens DESC", modelExpr)
 
 	rows, err := r.sql.QueryContext(ctx, query, args...)
 	if err != nil {
@@ -3021,7 +3043,7 @@ func (r *usageLogRepository) GetUserBreakdownStats(ctx context.Context, startTim
 		args = append(args, dim.GroupID)
 	}
 	if dim.Model != "" {
-		query += fmt.Sprintf(" AND ul.model = $%d", len(args)+1)
+		query += fmt.Sprintf(" AND %s = $%d", resolveModelDimensionExpression(dim.ModelType), len(args)+1)
 		args = append(args, dim.Model)
 	}
 	if dim.Endpoint != "" {
@@ -3102,6 +3124,18 @@ func (r *usageLogRepository) GetAllGroupUsageSummary(ctx context.Context, todayS
 	return results, nil
 }
 
+// resolveModelDimensionExpression maps model source type to a safe SQL expression.
+func resolveModelDimensionExpression(modelType string) string {
+	switch usagestats.NormalizeModelSource(modelType) {
+	case usagestats.ModelSourceUpstream:
+		return "COALESCE(NULLIF(TRIM(upstream_model), ''), model)"
+	case usagestats.ModelSourceMapping:
+		return "(model || ' -> ' || COALESCE(NULLIF(TRIM(upstream_model), ''), model))"
+	default:
+		return "model"
+	}
+}
+
 // resolveEndpointColumn maps endpoint type to the corresponding DB column name.
 func resolveEndpointColumn(endpointType string) string {
 	switch endpointType {
@@ -3854,6 +3888,7 @@ func scanUsageLog(scanner interface{ Scan(...any) error }) (*service.UsageLog, e
 		accountID             int64
 		requestID             sql.NullString
 		model                 string
+		upstreamModel         sql.NullString
 		groupID               sql.NullInt64
 		subscriptionID        sql.NullInt64
 		inputTokens           int
@@ -3896,6 +3931,7 @@ func scanUsageLog(scanner interface{ Scan(...any) error }) (*service.UsageLog, e
 		&accountID,
 		&requestID,
 		&model,
+		&upstreamModel,
 		&groupID,
 		&subscriptionID,
 		&inputTokens,
@@ -4008,6 +4044,9 @@ func scanUsageLog(scanner interface{ Scan(...any) error }) (*service.UsageLog, e
 	if upstreamEndpoint.Valid {
 		log.UpstreamEndpoint = &upstreamEndpoint.String
 	}
+	if upstreamModel.Valid {
+		log.UpstreamModel = &upstreamModel.String
+	}
 
 	return log, nil
 }

backend/internal/repository/usage_log_repo_breakdown_test.go

@@ -5,6 +5,7 @@ package repository
 import (
 	"testing"
 
+	"github.com/Wei-Shaw/sub2api/internal/pkg/usagestats"
 	"github.com/stretchr/testify/require"
 )
 
@@ -16,8 +17,8 @@ func TestResolveEndpointColumn(t *testing.T) {
 		{"inbound", "ul.inbound_endpoint"},
 		{"upstream", "ul.upstream_endpoint"},
 		{"path", "ul.inbound_endpoint || ' -> ' || ul.upstream_endpoint"},
-		{"", "ul.inbound_endpoint"},           // default
-		{"unknown", "ul.inbound_endpoint"},     // fallback
+		{"", "ul.inbound_endpoint"},        // default
+		{"unknown", "ul.inbound_endpoint"}, // fallback
 	}
 
 	for _, tc := range tests {
@@ -27,3 +28,23 @@ func TestResolveEndpointColumn(t *testing.T) {
 		})
 	}
 }
+
+func TestResolveModelDimensionExpression(t *testing.T) {
+	tests := []struct {
+		modelType string
+		want      string
+	}{
+		{usagestats.ModelSourceRequested, "model"},
+		{usagestats.ModelSourceUpstream, "COALESCE(NULLIF(TRIM(upstream_model), ''), model)"},
+		{usagestats.ModelSourceMapping, "(model || ' -> ' || COALESCE(NULLIF(TRIM(upstream_model), ''), model))"},
+		{"", "model"},
+		{"invalid", "model"},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.modelType, func(t *testing.T) {
+			got := resolveModelDimensionExpression(tc.modelType)
+			require.Equal(t, tc.want, got)
+		})
+	}
+}

backend/internal/repository/usage_log_repo_request_type_test.go

@@ -44,6 +44,7 @@ func TestUsageLogRepositoryCreateSyncRequestTypeAndLegacyFields(t *testing.T) {
 			log.AccountID,
 			log.RequestID,
 			log.Model,
+			sqlmock.AnyArg(), // upstream_model
 			sqlmock.AnyArg(), // group_id
 			sqlmock.AnyArg(), // subscription_id
 			log.InputTokens,
@@ -116,6 +117,7 @@ func TestUsageLogRepositoryCreate_PersistsServiceTier(t *testing.T) {
 			log.Model,
 			sqlmock.AnyArg(),
 			sqlmock.AnyArg(),
+			sqlmock.AnyArg(),
 			log.InputTokens,
 			log.OutputTokens,
 			log.CacheCreationTokens,
@@ -353,6 +355,7 @@ func TestScanUsageLogRequestTypeAndLegacyFallback(t *testing.T) {
 			int64(30), // account_id
 			sql.NullString{Valid: true, String: "req-1"},
 			"gpt-5",           // model
+			sql.NullString{},  // upstream_model
 			sql.NullInt64{},   // group_id
 			sql.NullInt64{},   // subscription_id
 			1,                 // input_tokens
@@ -404,6 +407,7 @@ func TestScanUsageLogRequestTypeAndLegacyFallback(t *testing.T) {
 			int64(31),
 			sql.NullString{Valid: true, String: "req-2"},
 			"gpt-5",
+			sql.NullString{},
 			sql.NullInt64{},
 			sql.NullInt64{},
 			1, 2, 3, 4, 5, 6,
@@ -445,6 +449,7 @@ func TestScanUsageLogRequestTypeAndLegacyFallback(t *testing.T) {
 			int64(32),
 			sql.NullString{Valid: true, String: "req-3"},
 			"gpt-5.4",
+			sql.NullString{},
 			sql.NullInt64{},
 			sql.NullInt64{},
 			1, 2, 3, 4, 5, 6,

backend/internal/server/routes/admin.go

@@ -402,6 +402,9 @@ func registerSettingsRoutes(admin *gin.RouterGroup, h *handler.Handlers) {
 		adminSettings.GET("/admin-api-key", h.Admin.Setting.GetAdminAPIKey)
 		adminSettings.POST("/admin-api-key/regenerate", h.Admin.Setting.RegenerateAdminAPIKey)
 		adminSettings.DELETE("/admin-api-key", h.Admin.Setting.DeleteAdminAPIKey)
+		// 529过载冷却配置
+		adminSettings.GET("/overload-cooldown", h.Admin.Setting.GetOverloadCooldownSettings)
+		adminSettings.PUT("/overload-cooldown", h.Admin.Setting.UpdateOverloadCooldownSettings)
 		// 流超时处理配置
 		adminSettings.GET("/stream-timeout", h.Admin.Setting.GetStreamTimeoutSettings)
 		adminSettings.PUT("/stream-timeout", h.Admin.Setting.UpdateStreamTimeoutSettings)

backend/internal/service/account_service.go

@@ -14,6 +14,8 @@ var (
 	ErrAccountNilInput = infraerrors.BadRequest("ACCOUNT_NIL_INPUT", "account input cannot be nil")
 )
 
+const AccountListGroupUngrouped int64 = -1
+
 type AccountRepository interface {
 	Create(ctx context.Context, account *Account) error
 	GetByID(ctx context.Context, id int64) (*Account, error)

backend/internal/service/account_test_service.go

@@ -308,7 +308,14 @@ func (s *AccountTestService) testClaudeAccountConnection(c *gin.Context, account
 
 	if resp.StatusCode != http.StatusOK {
 		body, _ := io.ReadAll(resp.Body)
-		return s.sendErrorAndEnd(c, fmt.Sprintf("API returned %d: %s", resp.StatusCode, string(body)))
+		errMsg := fmt.Sprintf("API returned %d: %s", resp.StatusCode, string(body))
+
+		// 403 表示账号被上游封禁，标记为 error 状态
+		if resp.StatusCode == http.StatusForbidden {
+			_ = s.accountRepo.SetError(ctx, account.ID, errMsg)
+		}
+
+		return s.sendErrorAndEnd(c, errMsg)
 	}
 
 	// Process SSE stream

backend/internal/service/account_usage_service.go

@@ -177,6 +177,7 @@ type AICredit struct {
 
 // UsageInfo 账号使用量信息
 type UsageInfo struct {
+	Source             string         `json:"source,omitempty"`               // "passive" or "active"
 	UpdatedAt          *time.Time     `json:"updated_at,omitempty"`           // 更新时间
 	FiveHour           *UsageProgress `json:"five_hour"`                      // 5小时窗口
 	SevenDay           *UsageProgress `json:"seven_day,omitempty"`            // 7天窗口
@@ -393,6 +394,9 @@ func (s *AccountUsageService) GetUsage(ctx context.Context, accountID int64) (*U
 		// 4. 添加窗口统计（有独立缓存，1 分钟）
 		s.addWindowStats(ctx, account, usage)
 
+		// 5. 将主动查询结果同步到被动缓存，下次 passive 加载即为最新值
+		s.syncActiveToPassive(ctx, account.ID, usage)
+
 		s.tryClearRecoverableAccountError(ctx, account)
 		return usage, nil
 	}
@@ -409,6 +413,81 @@ func (s *AccountUsageService) GetUsage(ctx context.Context, accountID int64) (*U
 	return nil, fmt.Errorf("account type %s does not support usage query", account.Type)
 }
 
+// GetPassiveUsage 从 Account.Extra 中的被动采样数据构建 UsageInfo，不调用外部 API。
+// 仅适用于 Anthropic OAuth / SetupToken 账号。
+func (s *AccountUsageService) GetPassiveUsage(ctx context.Context, accountID int64) (*UsageInfo, error) {
+	account, err := s.accountRepo.GetByID(ctx, accountID)
+	if err != nil {
+		return nil, fmt.Errorf("get account failed: %w", err)
+	}
+
+	if !account.IsAnthropicOAuthOrSetupToken() {
+		return nil, fmt.Errorf("passive usage only supported for Anthropic OAuth/SetupToken accounts")
+	}
+
+	// 复用 estimateSetupTokenUsage 构建 5h 窗口（OAuth 和 SetupToken 逻辑一致）
+	info := s.estimateSetupTokenUsage(account)
+	info.Source = "passive"
+
+	// 设置采样时间
+	if raw, ok := account.Extra["passive_usage_sampled_at"]; ok {
+		if str, ok := raw.(string); ok {
+			if t, err := time.Parse(time.RFC3339, str); err == nil {
+				info.UpdatedAt = &t
+			}
+		}
+	}
+
+	// 构建 7d 窗口（从被动采样数据）
+	util7d := parseExtraFloat64(account.Extra["passive_usage_7d_utilization"])
+	reset7dRaw := parseExtraFloat64(account.Extra["passive_usage_7d_reset"])
+	if util7d > 0 || reset7dRaw > 0 {
+		var resetAt *time.Time
+		var remaining int
+		if reset7dRaw > 0 {
+			t := time.Unix(int64(reset7dRaw), 0)
+			resetAt = &t
+			remaining = int(time.Until(t).Seconds())
+			if remaining < 0 {
+				remaining = 0
+			}
+		}
+		info.SevenDay = &UsageProgress{
+			Utilization:      util7d * 100,
+			ResetsAt:         resetAt,
+			RemainingSeconds: remaining,
+		}
+	}
+
+	// 添加窗口统计
+	s.addWindowStats(ctx, account, info)
+
+	return info, nil
+}
+
+// syncActiveToPassive 将主动查询的最新数据回写到 Extra 被动缓存，
+// 这样下次被动加载时能看到最新值。
+func (s *AccountUsageService) syncActiveToPassive(ctx context.Context, accountID int64, usage *UsageInfo) {
+	extraUpdates := make(map[string]any, 4)
+
+	if usage.FiveHour != nil {
+		extraUpdates["session_window_utilization"] = usage.FiveHour.Utilization / 100
+	}
+	if usage.SevenDay != nil {
+		extraUpdates["passive_usage_7d_utilization"] = usage.SevenDay.Utilization / 100
+		if usage.SevenDay.ResetsAt != nil {
+			extraUpdates["passive_usage_7d_reset"] = usage.SevenDay.ResetsAt.Unix()
+		}
+	}
+
+	if len(extraUpdates) > 0 {
+		extraUpdates["passive_usage_sampled_at"] = time.Now().UTC().Format(time.RFC3339)
+		if err := s.accountRepo.UpdateExtra(ctx, accountID, extraUpdates); err != nil {
+			slog.Warn("sync_active_to_passive_failed", "account_id", accountID, "error", err)
+		}
+	}
+}
+
 func (s *AccountUsageService) getOpenAIUsage(ctx context.Context, account *Account) (*UsageInfo, error) {
 	now := time.Now()
 	usage := &UsageInfo{UpdatedAt: &now}

backend/internal/service/antigravity_model_mapping_test.go

@@ -57,16 +57,16 @@ func TestAntigravityGatewayService_GetMappedModel(t *testing.T) {
 			expected:       "claude-opus-4-6-thinking",
 		},
 		{
-			name:           "默认映射 - claude-haiku-4-5 → claude-sonnet-4-5",
+			name:           "默认映射 - claude-haiku-4-5 → claude-sonnet-4-6",
 			requestedModel: "claude-haiku-4-5",
 			accountMapping: nil,
-			expected:       "claude-sonnet-4-5",
+			expected:       "claude-sonnet-4-6",
 		},
 		{
-			name:           "默认映射 - claude-haiku-4-5-20251001 → claude-sonnet-4-5",
+			name:           "默认映射 - claude-haiku-4-5-20251001 → claude-sonnet-4-6",
 			requestedModel: "claude-haiku-4-5-20251001",
 			accountMapping: nil,
-			expected:       "claude-sonnet-4-5",
+			expected:       "claude-sonnet-4-6",
 		},
 		{
 			name:           "默认映射 - claude-sonnet-4-5-20250929 → claude-sonnet-4-5",

backend/internal/service/dashboard_service.go

@@ -140,6 +140,27 @@ func (s *DashboardService) GetModelStatsWithFilters(ctx context.Context, startTi
 	return stats, nil
 }
 
+func (s *DashboardService) GetModelStatsWithFiltersBySource(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8, modelSource string) ([]usagestats.ModelStat, error) {
+	normalizedSource := usagestats.NormalizeModelSource(modelSource)
+	if normalizedSource == usagestats.ModelSourceRequested {
+		return s.GetModelStatsWithFilters(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+	}
+
+	type modelStatsBySourceRepo interface {
+		GetModelStatsWithFiltersBySource(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8, source string) ([]usagestats.ModelStat, error)
+	}
+
+	if sourceRepo, ok := s.usageRepo.(modelStatsBySourceRepo); ok {
+		stats, err := sourceRepo.GetModelStatsWithFiltersBySource(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType, normalizedSource)
+		if err != nil {
+			return nil, fmt.Errorf("get model stats with filters by source: %w", err)
+		}
+		return stats, nil
+	}
+
+	return s.GetModelStatsWithFilters(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
+}
+
 func (s *DashboardService) GetGroupStatsWithFilters(ctx context.Context, startTime, endTime time.Time, userID, apiKeyID, accountID, groupID int64, requestType *int16, stream *bool, billingType *int8) ([]usagestats.GroupStat, error) {
 	stats, err := s.usageRepo.GetGroupStatsWithFilters(ctx, startTime, endTime, userID, apiKeyID, accountID, groupID, requestType, stream, billingType)
 	if err != nil {

backend/internal/service/domain_constants.go

@@ -170,6 +170,13 @@ const (
 	// SettingKeyOpsRuntimeLogConfig stores JSON config for runtime log settings.
 	SettingKeyOpsRuntimeLogConfig = "ops_runtime_log_config"
 
+	// =========================
+	// Overload Cooldown (529)
+	// =========================
+
+	// SettingKeyOverloadCooldownSettings stores JSON config for 529 overload cooldown handling.
+	SettingKeyOverloadCooldownSettings = "overload_cooldown_settings"
+
 	// =========================
 	// Stream Timeout Handling
 	// =========================

backend/internal/service/gateway_anthropic_apikey_passthrough_test.go

@@ -688,6 +688,83 @@ func TestGatewayService_AnthropicOAuth_NotAffectedByAPIKeyPassthroughToggle(t *t
 	require.Contains(t, req.Header.Get("anthropic-beta"), claude.BetaOAuth, "OAuth 链路仍应按原逻辑补齐 oauth beta")
 }
 
+func TestGatewayService_AnthropicOAuth_ForwardPreservesBillingHeaderSystemBlock(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	tests := []struct {
+		name string
+		body string
+	}{
+		{
+			name: "system array",
+			body: `{"model":"claude-3-5-sonnet-latest","system":[{"type":"text","text":"x-anthropic-billing-header keep"}],"messages":[{"role":"user","content":[{"type":"text","text":"hello"}]}]}`,
+		},
+		{
+			name: "system string",
+			body: `{"model":"claude-3-5-sonnet-latest","system":"x-anthropic-billing-header keep","messages":[{"role":"user","content":[{"type":"text","text":"hello"}]}]}`,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodPost, "/v1/messages", nil)
+
+			parsed, err := ParseGatewayRequest([]byte(tt.body), PlatformAnthropic)
+			require.NoError(t, err)
+
+			upstream := &anthropicHTTPUpstreamRecorder{
+				resp: &http.Response{
+					StatusCode: http.StatusOK,
+					Header: http.Header{
+						"Content-Type": []string{"application/json"},
+						"x-request-id": []string{"rid-oauth-preserve"},
+					},
+					Body: io.NopCloser(strings.NewReader(`{"id":"msg_1","type":"message","role":"assistant","model":"claude-3-5-sonnet-20241022","content":[{"type":"text","text":"ok"}],"usage":{"input_tokens":12,"output_tokens":7}}`)),
+				},
+			}
+
+			cfg := &config.Config{
+				Gateway: config.GatewayConfig{
+					MaxLineSize: defaultMaxLineSize,
+				},
+			}
+			svc := &GatewayService{
+				cfg:                  cfg,
+				responseHeaderFilter: compileResponseHeaderFilter(cfg),
+				httpUpstream:         upstream,
+				rateLimitService:     &RateLimitService{},
+				deferredService:      &DeferredService{},
+			}
+
+			account := &Account{
+				ID:          301,
+				Name:        "anthropic-oauth-preserve",
+				Platform:    PlatformAnthropic,
+				Type:        AccountTypeOAuth,
+				Concurrency: 1,
+				Credentials: map[string]any{
+					"access_token": "oauth-token",
+				},
+				Status:      StatusActive,
+				Schedulable: true,
+			}
+
+			result, err := svc.Forward(context.Background(), c, account, parsed)
+			require.NoError(t, err)
+			require.NotNil(t, result)
+			require.NotNil(t, upstream.lastReq)
+			require.Equal(t, "Bearer oauth-token", upstream.lastReq.Header.Get("authorization"))
+			require.Contains(t, upstream.lastReq.Header.Get("anthropic-beta"), claude.BetaOAuth)
+
+			system := gjson.GetBytes(upstream.lastBody, "system")
+			require.True(t, system.Exists())
+			require.Contains(t, system.Raw, "x-anthropic-billing-header keep")
+		})
+	}
+}
+
 func TestGatewayService_AnthropicAPIKeyPassthrough_StreamingStillCollectsUsageAfterClientDisconnect(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
@@ -788,7 +865,7 @@ func TestGatewayService_AnthropicAPIKeyPassthrough_ForwardDirect_NonStreamingSuc
 		rateLimitService: &RateLimitService{},
 	}
 
-	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, newAnthropicAPIKeyAccountForTest(), body, "claude-3-5-sonnet-latest", false, time.Now())
+	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, newAnthropicAPIKeyAccountForTest(), body, "claude-3-5-sonnet-latest", "claude-3-5-sonnet-latest", false, time.Now())
 	require.NoError(t, err)
 	require.NotNil(t, result)
 	require.Equal(t, 12, result.Usage.InputTokens)
@@ -815,7 +892,7 @@ func TestGatewayService_AnthropicAPIKeyPassthrough_ForwardDirect_InvalidTokenTyp
 	}
 	svc := &GatewayService{}
 
-	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, account, []byte(`{}`), "claude-3-5-sonnet-latest", false, time.Now())
+	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, account, []byte(`{}`), "claude-3-5-sonnet-latest", "claude-3-5-sonnet-latest", false, time.Now())
 	require.Nil(t, result)
 	require.Error(t, err)
 	require.Contains(t, err.Error(), "requires apikey token")
@@ -840,7 +917,7 @@ func TestGatewayService_AnthropicAPIKeyPassthrough_ForwardDirect_UpstreamRequest
 	}
 	account := newAnthropicAPIKeyAccountForTest()
 
-	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, account, []byte(`{"model":"x"}`), "x", false, time.Now())
+	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, account, []byte(`{"model":"x"}`), "x", "x", false, time.Now())
 	require.Nil(t, result)
 	require.Error(t, err)
 	require.Contains(t, err.Error(), "upstream request failed")
@@ -873,7 +950,7 @@ func TestGatewayService_AnthropicAPIKeyPassthrough_ForwardDirect_EmptyResponseBo
 		httpUpstream: upstream,
 	}
 
-	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, newAnthropicAPIKeyAccountForTest(), []byte(`{"model":"x"}`), "x", false, time.Now())
+	result, err := svc.forwardAnthropicAPIKeyPassthrough(context.Background(), c, newAnthropicAPIKeyAccountForTest(), []byte(`{"model":"x"}`), "x", "x", false, time.Now())
 	require.Nil(t, result)
 	require.Error(t, err)
 	require.Contains(t, err.Error(), "empty response")

backend/internal/service/gateway_body_order_test.go

@@ -0,0 +1,72 @@
+package service
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
+	"github.com/stretchr/testify/require"
+)
+
+func assertJSONTokenOrder(t *testing.T, body string, tokens ...string) {
+	t.Helper()
+
+	last := -1
+	for _, token := range tokens {
+		pos := strings.Index(body, token)
+		require.NotEqualf(t, -1, pos, "missing token %s in body %s", token, body)
+		require.Greaterf(t, pos, last, "token %s should appear after previous tokens in body %s", token, body)
+		last = pos
+	}
+}
+
+func TestReplaceModelInBody_PreservesTopLevelFieldOrder(t *testing.T) {
+	svc := &GatewayService{}
+	body := []byte(`{"alpha":1,"model":"claude-3-5-sonnet-latest","messages":[],"omega":2}`)
+
+	result := svc.replaceModelInBody(body, "claude-3-5-sonnet-20241022")
+	resultStr := string(result)
+
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"model"`, `"messages"`, `"omega"`)
+	require.Contains(t, resultStr, `"model":"claude-3-5-sonnet-20241022"`)
+}
+
+func TestNormalizeClaudeOAuthRequestBody_PreservesTopLevelFieldOrder(t *testing.T) {
+	body := []byte(`{"alpha":1,"model":"claude-3-5-sonnet-latest","temperature":0.2,"system":"You are OpenCode, the best coding agent on the planet.","messages":[],"tool_choice":{"type":"auto"},"omega":2}`)
+
+	result, modelID := normalizeClaudeOAuthRequestBody(body, "claude-3-5-sonnet-latest", claudeOAuthNormalizeOptions{
+		injectMetadata: true,
+		metadataUserID: "user-1",
+	})
+	resultStr := string(result)
+
+	require.Equal(t, claude.NormalizeModelID("claude-3-5-sonnet-latest"), modelID)
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"model"`, `"system"`, `"messages"`, `"omega"`, `"tools"`, `"metadata"`)
+	require.NotContains(t, resultStr, `"temperature"`)
+	require.NotContains(t, resultStr, `"tool_choice"`)
+	require.Contains(t, resultStr, `"system":"`+claudeCodeSystemPrompt+`"`)
+	require.Contains(t, resultStr, `"tools":[]`)
+	require.Contains(t, resultStr, `"metadata":{"user_id":"user-1"}`)
+}
+
+func TestInjectClaudeCodePrompt_PreservesFieldOrder(t *testing.T) {
+	body := []byte(`{"alpha":1,"system":[{"id":"block-1","type":"text","text":"Custom"}],"messages":[],"omega":2}`)
+
+	result := injectClaudeCodePrompt(body, []any{
+		map[string]any{"id": "block-1", "type": "text", "text": "Custom"},
+	})
+	resultStr := string(result)
+
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"system"`, `"messages"`, `"omega"`)
+	require.Contains(t, resultStr, `{"id":"block-1","type":"text","text":"`+claudeCodeSystemPrompt+`\n\nCustom"}`)
+}
+
+func TestEnforceCacheControlLimit_PreservesTopLevelFieldOrder(t *testing.T) {
+	body := []byte(`{"alpha":1,"system":[{"type":"text","text":"s1","cache_control":{"type":"ephemeral"}},{"type":"text","text":"s2","cache_control":{"type":"ephemeral"}}],"messages":[{"role":"user","content":[{"type":"text","text":"m1","cache_control":{"type":"ephemeral"}},{"type":"text","text":"m2","cache_control":{"type":"ephemeral"}},{"type":"text","text":"m3","cache_control":{"type":"ephemeral"}}]}],"omega":2}`)
+
+	result := enforceCacheControlLimit(body)
+	resultStr := string(result)
+
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"system"`, `"messages"`, `"omega"`)
+	require.Equal(t, 4, strings.Count(resultStr, `"cache_control"`))
+}

backend/internal/service/gateway_debug_env_test.go

@@ -0,0 +1,34 @@
+package service
+
+import "testing"
+
+func TestDebugGatewayBodyLoggingEnabled(t *testing.T) {
+	t.Run("default disabled", func(t *testing.T) {
+		t.Setenv(debugGatewayBodyEnv, "")
+		if debugGatewayBodyLoggingEnabled() {
+			t.Fatalf("expected debug gateway body logging to be disabled by default")
+		}
+	})
+
+	t.Run("enabled with true-like values", func(t *testing.T) {
+		for _, value := range []string{"1", "true", "TRUE", "yes", "on"} {
+			t.Run(value, func(t *testing.T) {
+				t.Setenv(debugGatewayBodyEnv, value)
+				if !debugGatewayBodyLoggingEnabled() {
+					t.Fatalf("expected debug gateway body logging to be enabled for %q", value)
+				}
+			})
+		}
+	})
+
+	t.Run("disabled with other values", func(t *testing.T) {
+		for _, value := range []string{"0", "false", "off", "debug"} {
+			t.Run(value, func(t *testing.T) {
+				t.Setenv(debugGatewayBodyEnv, value)
+				if debugGatewayBodyLoggingEnabled() {
+					t.Fatalf("expected debug gateway body logging to be disabled for %q", value)
+				}
+			})
+		}
+	})
+}

backend/internal/service/gateway_request.go

@@ -28,6 +28,12 @@ var (
 	patternEmptyContentSpaced = []byte(`"content": []`)
 	patternEmptyContentSp1    = []byte(`"content" : []`)
 	patternEmptyContentSp2    = []byte(`"content" :[]`)
+
+	// Fast-path patterns for empty text blocks: {"type":"text","text":""}
+	patternEmptyText       = []byte(`"text":""`)
+	patternEmptyTextSpaced = []byte(`"text": ""`)
+	patternEmptyTextSp1    = []byte(`"text" : ""`)
+	patternEmptyTextSp2    = []byte(`"text" :""`)
 )
 
 // SessionContext 粘性会话上下文，用于区分不同来源的请求。
@@ -233,15 +239,22 @@ func FilterThinkingBlocksForRetry(body []byte) []byte {
 		bytes.Contains(body, patternThinkingField) ||
 		bytes.Contains(body, patternThinkingFieldSpaced)
 
-	// Also check for empty content arrays that need fixing.
+	// Also check for empty content arrays and empty text blocks that need fixing.
 	// Note: This is a heuristic check; the actual empty content handling is done below.
 	hasEmptyContent := bytes.Contains(body, patternEmptyContent) ||
 		bytes.Contains(body, patternEmptyContentSpaced) ||
 		bytes.Contains(body, patternEmptyContentSp1) ||
 		bytes.Contains(body, patternEmptyContentSp2)
 
+	// Check for empty text blocks: {"type":"text","text":""}
+	// These cause upstream 400: "text content blocks must be non-empty"
+	hasEmptyTextBlock := bytes.Contains(body, patternEmptyText) ||
+		bytes.Contains(body, patternEmptyTextSpaced) ||
+		bytes.Contains(body, patternEmptyTextSp1) ||
+		bytes.Contains(body, patternEmptyTextSp2)
+
 	// Fast path: nothing to process
-	if !hasThinkingContent && !hasEmptyContent {
+	if !hasThinkingContent && !hasEmptyContent && !hasEmptyTextBlock {
 		return body
 	}
 
@@ -260,7 +273,7 @@ func FilterThinkingBlocksForRetry(body []byte) []byte {
 		bytes.Contains(body, patternTypeRedactedThinking) ||
 		bytes.Contains(body, patternTypeRedactedSpaced) ||
 		bytes.Contains(body, patternThinkingFieldSpaced)
-	if !hasEmptyContent && !containsThinkingBlocks {
+	if !hasEmptyContent && !hasEmptyTextBlock && !containsThinkingBlocks {
 		if topThinking := gjson.Get(jsonStr, "thinking"); topThinking.Exists() {
 			if out, err := sjson.DeleteBytes(body, "thinking"); err == nil {
 				out = removeThinkingDependentContextStrategies(out)
@@ -320,6 +333,16 @@ func FilterThinkingBlocksForRetry(body []byte) []byte {
 
 			blockType, _ := blockMap["type"].(string)
 
+			// Strip empty text blocks: {"type":"text","text":""}
+			// Upstream rejects these with 400: "text content blocks must be non-empty"
+			if blockType == "text" {
+				if txt, _ := blockMap["text"].(string); txt == "" {
+					modifiedThisMsg = true
+					ensureNewContent(bi)
+					continue
+				}
+			}
+
 			// Convert thinking blocks to text (preserve content) and drop redacted_thinking.
 			switch blockType {
 			case "thinking":

backend/internal/service/gateway_request_test.go

@@ -404,6 +404,51 @@ func TestFilterThinkingBlocksForRetry_EmptyContentGetsPlaceholder(t *testing.T)
 	require.NotEmpty(t, content0["text"])
 }
 
+func TestFilterThinkingBlocksForRetry_StripsEmptyTextBlocks(t *testing.T) {
+	// Empty text blocks cause upstream 400: "text content blocks must be non-empty"
+	input := []byte(`{
+		"messages":[
+			{"role":"user","content":[{"type":"text","text":"hello"},{"type":"text","text":""}]},
+			{"role":"assistant","content":[{"type":"text","text":""}]}
+		]
+	}`)
+
+	out := FilterThinkingBlocksForRetry(input)
+
+	var req map[string]any
+	require.NoError(t, json.Unmarshal(out, &req))
+	msgs, ok := req["messages"].([]any)
+	require.True(t, ok)
+
+	// First message: empty text block stripped, "hello" preserved
+	msg0 := msgs[0].(map[string]any)
+	content0 := msg0["content"].([]any)
+	require.Len(t, content0, 1)
+	require.Equal(t, "hello", content0[0].(map[string]any)["text"])
+
+	// Second message: only had empty text block → gets placeholder
+	msg1 := msgs[1].(map[string]any)
+	content1 := msg1["content"].([]any)
+	require.Len(t, content1, 1)
+	block1 := content1[0].(map[string]any)
+	require.Equal(t, "text", block1["type"])
+	require.NotEmpty(t, block1["text"])
+}
+
+func TestFilterThinkingBlocksForRetry_PreservesNonEmptyTextBlocks(t *testing.T) {
+	// Non-empty text blocks should pass through unchanged
+	input := []byte(`{
+		"messages":[
+			{"role":"user","content":[{"type":"text","text":"hello"},{"type":"text","text":"world"}]}
+		]
+	}`)
+
+	out := FilterThinkingBlocksForRetry(input)
+
+	// Fast path: no thinking content, no empty content, no empty text blocks → unchanged
+	require.Equal(t, input, out)
+}
+
 func TestFilterSignatureSensitiveBlocksForRetry_DowngradesTools(t *testing.T) {
 	input := []byte(`{
 		"thinking":{"type":"enabled","budget_tokens":1024},

backend/internal/service/identity_service.go

@@ -5,7 +5,6 @@ import (
 	"crypto/rand"
 	"crypto/sha256"
 	"encoding/hex"
-	"encoding/json"
 	"fmt"
 	"log/slog"
 	"net/http"
@@ -15,6 +14,8 @@ import (
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/tidwall/gjson"
+	"github.com/tidwall/sjson"
 )
 
 // 预编译正则表达式（避免每次调用重新编译）
@@ -215,25 +216,20 @@ func (s *IdentityService) RewriteUserID(body []byte, accountID int64, accountUUI
 		return body, nil
 	}
 
-	// 使用 RawMessage 保留其他字段的原始字节
-	var reqMap map[string]json.RawMessage
-	if err := json.Unmarshal(body, &reqMap); err != nil {
+	metadata := gjson.GetBytes(body, "metadata")
+	if !metadata.Exists() || metadata.Type == gjson.Null {
+		return body, nil
+	}
+	if !strings.HasPrefix(strings.TrimSpace(metadata.Raw), "{") {
 		return body, nil
 	}
 
-	// 解析 metadata 字段
-	metadataRaw, ok := reqMap["metadata"]
-	if !ok {
+	userIDResult := metadata.Get("user_id")
+	if !userIDResult.Exists() || userIDResult.Type != gjson.String {
 		return body, nil
 	}
-
-	var metadata map[string]any
-	if err := json.Unmarshal(metadataRaw, &metadata); err != nil {
-		return body, nil
-	}
-
-	userID, ok := metadata["user_id"].(string)
-	if !ok || userID == "" {
+	userID := userIDResult.String()
+	if userID == "" {
 		return body, nil
 	}
 
@@ -252,17 +248,15 @@ func (s *IdentityService) RewriteUserID(body []byte, accountID int64, accountUUI
 	// 根据客户端版本选择输出格式
 	version := ExtractCLIVersion(fingerprintUA)
 	newUserID := FormatMetadataUserID(cachedClientID, accountUUID, newSessionHash, version)
+	if newUserID == userID {
+		return body, nil
+	}
 
-	metadata["user_id"] = newUserID
-
-	// 只重新序列化 metadata 字段
-	newMetadataRaw, err := json.Marshal(metadata)
+	newBody, err := sjson.SetBytes(body, "metadata.user_id", newUserID)
 	if err != nil {
 		return body, nil
 	}
-	reqMap["metadata"] = newMetadataRaw
-
-	return json.Marshal(reqMap)
+	return newBody, nil
 }
 
 // RewriteUserIDWithMasking 重写body中的metadata.user_id，支持会话ID伪装
@@ -283,25 +277,20 @@ func (s *IdentityService) RewriteUserIDWithMasking(ctx context.Context, body []b
 		return newBody, nil
 	}
 
-	// 使用 RawMessage 保留其他字段的原始字节
-	var reqMap map[string]json.RawMessage
-	if err := json.Unmarshal(newBody, &reqMap); err != nil {
+	metadata := gjson.GetBytes(newBody, "metadata")
+	if !metadata.Exists() || metadata.Type == gjson.Null {
+		return newBody, nil
+	}
+	if !strings.HasPrefix(strings.TrimSpace(metadata.Raw), "{") {
 		return newBody, nil
 	}
 
-	// 解析 metadata 字段
-	metadataRaw, ok := reqMap["metadata"]
-	if !ok {
+	userIDResult := metadata.Get("user_id")
+	if !userIDResult.Exists() || userIDResult.Type != gjson.String {
 		return newBody, nil
 	}
-
-	var metadata map[string]any
-	if err := json.Unmarshal(metadataRaw, &metadata); err != nil {
-		return newBody, nil
-	}
-
-	userID, ok := metadata["user_id"].(string)
-	if !ok || userID == "" {
+	userID := userIDResult.String()
+	if userID == "" {
 		return newBody, nil
 	}
 
@@ -339,16 +328,15 @@ func (s *IdentityService) RewriteUserIDWithMasking(ctx context.Context, body []b
 		"after", newUserID,
 	)
 
-	metadata["user_id"] = newUserID
-
-	// 只重新序列化 metadata 字段
-	newMetadataRaw, marshalErr := json.Marshal(metadata)
-	if marshalErr != nil {
+	if newUserID == userID {
 		return newBody, nil
 	}
-	reqMap["metadata"] = newMetadataRaw
 
-	return json.Marshal(reqMap)
+	maskedBody, setErr := sjson.SetBytes(newBody, "metadata.user_id", newUserID)
+	if setErr != nil {
+		return newBody, nil
+	}
+	return maskedBody, nil
 }
 
 // generateRandomUUID 生成随机 UUID v4 格式字符串

backend/internal/service/identity_service_order_test.go

@@ -0,0 +1,82 @@
+package service
+
+import (
+	"context"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+type identityCacheStub struct {
+	maskedSessionID string
+}
+
+func (s *identityCacheStub) GetFingerprint(_ context.Context, _ int64) (*Fingerprint, error) {
+	return nil, nil
+}
+func (s *identityCacheStub) SetFingerprint(_ context.Context, _ int64, _ *Fingerprint) error {
+	return nil
+}
+func (s *identityCacheStub) GetMaskedSessionID(_ context.Context, _ int64) (string, error) {
+	return s.maskedSessionID, nil
+}
+func (s *identityCacheStub) SetMaskedSessionID(_ context.Context, _ int64, sessionID string) error {
+	s.maskedSessionID = sessionID
+	return nil
+}
+
+func TestIdentityService_RewriteUserID_PreservesTopLevelFieldOrder(t *testing.T) {
+	cache := &identityCacheStub{}
+	svc := NewIdentityService(cache)
+
+	originalUserID := FormatMetadataUserID(
+		"d61f76d0730d2b920763648949bad5c79742155c27037fc77ac3f9805cb90169",
+		"",
+		"7578cf37-aaca-46e4-a45c-71285d9dbb83",
+		"2.1.78",
+	)
+	body := []byte(`{"alpha":1,"messages":[],"metadata":{"user_id":` + strconvQuote(originalUserID) + `},"max_tokens":64000,"thinking":{"type":"adaptive"},"output_config":{"effort":"high"},"stream":true}`)
+
+	result, err := svc.RewriteUserID(body, 123, "acc-uuid", "client-xyz", "claude-cli/2.1.78 (external, cli)")
+	require.NoError(t, err)
+	resultStr := string(result)
+
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"messages"`, `"metadata"`, `"max_tokens"`, `"thinking"`, `"output_config"`, `"stream"`)
+	require.NotContains(t, resultStr, originalUserID)
+	require.Contains(t, resultStr, `"metadata":{"user_id":"`)
+}
+
+func TestIdentityService_RewriteUserIDWithMasking_PreservesTopLevelFieldOrder(t *testing.T) {
+	cache := &identityCacheStub{maskedSessionID: "11111111-2222-4333-8444-555555555555"}
+	svc := NewIdentityService(cache)
+
+	originalUserID := FormatMetadataUserID(
+		"d61f76d0730d2b920763648949bad5c79742155c27037fc77ac3f9805cb90169",
+		"",
+		"7578cf37-aaca-46e4-a45c-71285d9dbb83",
+		"2.1.78",
+	)
+	body := []byte(`{"alpha":1,"messages":[],"metadata":{"user_id":` + strconvQuote(originalUserID) + `},"max_tokens":64000,"thinking":{"type":"adaptive"},"output_config":{"effort":"high"},"stream":true}`)
+
+	account := &Account{
+		ID:       123,
+		Platform: PlatformAnthropic,
+		Type:     AccountTypeOAuth,
+		Extra: map[string]any{
+			"session_id_masking_enabled": true,
+		},
+	}
+
+	result, err := svc.RewriteUserIDWithMasking(context.Background(), body, account, "acc-uuid", "client-xyz", "claude-cli/2.1.78 (external, cli)")
+	require.NoError(t, err)
+	resultStr := string(result)
+
+	assertJSONTokenOrder(t, resultStr, `"alpha"`, `"messages"`, `"metadata"`, `"max_tokens"`, `"thinking"`, `"output_config"`, `"stream"`)
+	require.Contains(t, resultStr, cache.maskedSessionID)
+	require.True(t, strings.Contains(resultStr, `"metadata":{"user_id":"`))
+}
+
+func strconvQuote(v string) string {
+	return `"` + strings.ReplaceAll(strings.ReplaceAll(v, `\`, `\\`), `"`, `\"`) + `"`
+}

backend/internal/service/openai_compat_prompt_cache_key.go

@@ -0,0 +1,81 @@
+package service
+
+import (
+	"encoding/json"
+	"strings"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+)
+
+const compatPromptCacheKeyPrefix = "compat_cc_"
+
+func shouldAutoInjectPromptCacheKeyForCompat(model string) bool {
+	switch normalizeCodexModel(strings.TrimSpace(model)) {
+	case "gpt-5.4", "gpt-5.3-codex":
+		return true
+	default:
+		return false
+	}
+}
+
+func deriveCompatPromptCacheKey(req *apicompat.ChatCompletionsRequest, mappedModel string) string {
+	if req == nil {
+		return ""
+	}
+
+	normalizedModel := normalizeCodexModel(strings.TrimSpace(mappedModel))
+	if normalizedModel == "" {
+		normalizedModel = normalizeCodexModel(strings.TrimSpace(req.Model))
+	}
+	if normalizedModel == "" {
+		normalizedModel = strings.TrimSpace(req.Model)
+	}
+
+	seedParts := []string{"model=" + normalizedModel}
+	if req.ReasoningEffort != "" {
+		seedParts = append(seedParts, "reasoning_effort="+strings.TrimSpace(req.ReasoningEffort))
+	}
+	if len(req.ToolChoice) > 0 {
+		seedParts = append(seedParts, "tool_choice="+normalizeCompatSeedJSON(req.ToolChoice))
+	}
+	if len(req.Tools) > 0 {
+		if raw, err := json.Marshal(req.Tools); err == nil {
+			seedParts = append(seedParts, "tools="+normalizeCompatSeedJSON(raw))
+		}
+	}
+	if len(req.Functions) > 0 {
+		if raw, err := json.Marshal(req.Functions); err == nil {
+			seedParts = append(seedParts, "functions="+normalizeCompatSeedJSON(raw))
+		}
+	}
+
+	firstUserCaptured := false
+	for _, msg := range req.Messages {
+		switch strings.TrimSpace(msg.Role) {
+		case "system":
+			seedParts = append(seedParts, "system="+normalizeCompatSeedJSON(msg.Content))
+		case "user":
+			if !firstUserCaptured {
+				seedParts = append(seedParts, "first_user="+normalizeCompatSeedJSON(msg.Content))
+				firstUserCaptured = true
+			}
+		}
+	}
+
+	return compatPromptCacheKeyPrefix + hashSensitiveValueForLog(strings.Join(seedParts, "|"))
+}
+
+func normalizeCompatSeedJSON(v json.RawMessage) string {
+	if len(v) == 0 {
+		return ""
+	}
+	var tmp any
+	if err := json.Unmarshal(v, &tmp); err != nil {
+		return string(v)
+	}
+	out, err := json.Marshal(tmp)
+	if err != nil {
+		return string(v)
+	}
+	return string(out)
+}

backend/internal/service/openai_compat_prompt_cache_key_test.go

@@ -0,0 +1,64 @@
+package service
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/stretchr/testify/require"
+)
+
+func mustRawJSON(t *testing.T, s string) json.RawMessage {
+	t.Helper()
+	return json.RawMessage(s)
+}
+
+func TestShouldAutoInjectPromptCacheKeyForCompat(t *testing.T) {
+	require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.4"))
+	require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.3"))
+	require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.3-codex"))
+	require.False(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-4o"))
+}
+
+func TestDeriveCompatPromptCacheKey_StableAcrossLaterTurns(t *testing.T) {
+	base := &apicompat.ChatCompletionsRequest{
+		Model: "gpt-5.4",
+		Messages: []apicompat.ChatMessage{
+			{Role: "system", Content: mustRawJSON(t, `"You are helpful."`)},
+			{Role: "user", Content: mustRawJSON(t, `"Hello"`)},
+		},
+	}
+	extended := &apicompat.ChatCompletionsRequest{
+		Model: "gpt-5.4",
+		Messages: []apicompat.ChatMessage{
+			{Role: "system", Content: mustRawJSON(t, `"You are helpful."`)},
+			{Role: "user", Content: mustRawJSON(t, `"Hello"`)},
+			{Role: "assistant", Content: mustRawJSON(t, `"Hi there!"`)},
+			{Role: "user", Content: mustRawJSON(t, `"How are you?"`)},
+		},
+	}
+
+	k1 := deriveCompatPromptCacheKey(base, "gpt-5.4")
+	k2 := deriveCompatPromptCacheKey(extended, "gpt-5.4")
+	require.Equal(t, k1, k2, "cache key should be stable across later turns")
+	require.NotEmpty(t, k1)
+}
+
+func TestDeriveCompatPromptCacheKey_DiffersAcrossSessions(t *testing.T) {
+	req1 := &apicompat.ChatCompletionsRequest{
+		Model: "gpt-5.4",
+		Messages: []apicompat.ChatMessage{
+			{Role: "user", Content: mustRawJSON(t, `"Question A"`)},
+		},
+	}
+	req2 := &apicompat.ChatCompletionsRequest{
+		Model: "gpt-5.4",
+		Messages: []apicompat.ChatMessage{
+			{Role: "user", Content: mustRawJSON(t, `"Question B"`)},
+		},
+	}
+
+	k1 := deriveCompatPromptCacheKey(req1, "gpt-5.4")
+	k2 := deriveCompatPromptCacheKey(req2, "gpt-5.4")
+	require.NotEqual(t, k1, k2, "different first user messages should yield different keys")
+}

backend/internal/service/openai_gateway_chat_completions.go

@@ -43,23 +43,38 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions(
 	clientStream := chatReq.Stream
 	includeUsage := chatReq.StreamOptions != nil && chatReq.StreamOptions.IncludeUsage
 
-	// 2. Convert to Responses and forward
+	// 2. Resolve model mapping early so compat prompt_cache_key injection can
+	// derive a stable seed from the final upstream model family.
+	mappedModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
+
+	promptCacheKey = strings.TrimSpace(promptCacheKey)
+	compatPromptCacheInjected := false
+	if promptCacheKey == "" && account.Type == AccountTypeOAuth && shouldAutoInjectPromptCacheKeyForCompat(mappedModel) {
+		promptCacheKey = deriveCompatPromptCacheKey(&chatReq, mappedModel)
+		compatPromptCacheInjected = promptCacheKey != ""
+	}
+
+	// 3. Convert to Responses and forward
 	// ChatCompletionsToResponses always sets Stream=true (upstream always streams).
 	responsesReq, err := apicompat.ChatCompletionsToResponses(&chatReq)
 	if err != nil {
 		return nil, fmt.Errorf("convert chat completions to responses: %w", err)
 	}
-
-	// 3. Model mapping
-	mappedModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
 	responsesReq.Model = mappedModel
 
-	logger.L().Debug("openai chat_completions: model mapping applied",
+	logFields := []zap.Field{
 		zap.Int64("account_id", account.ID),
 		zap.String("original_model", originalModel),
 		zap.String("mapped_model", mappedModel),
 		zap.Bool("stream", clientStream),
-	)
+	}
+	if compatPromptCacheInjected {
+		logFields = append(logFields,
+			zap.Bool("compat_prompt_cache_key_injected", true),
+			zap.String("compat_prompt_cache_key_sha256", hashSensitiveValueForLog(promptCacheKey)),
+		)
+	}
+	logger.L().Debug("openai chat_completions: model mapping applied", logFields...)
 
 	// 4. Marshal Responses request body, then apply OAuth codex transform
 	responsesBody, err := json.Marshal(responsesReq)
@@ -277,12 +292,13 @@ func (s *OpenAIGatewayService) handleChatBufferedStreamingResponse(
 	c.JSON(http.StatusOK, chatResp)
 
 	return &OpenAIForwardResult{
-		RequestID:    requestID,
-		Usage:        usage,
-		Model:        originalModel,
-		BillingModel: mappedModel,
-		Stream:       false,
-		Duration:     time.Since(startTime),
+		RequestID:     requestID,
+		Usage:         usage,
+		Model:         originalModel,
+		BillingModel:  mappedModel,
+		UpstreamModel: mappedModel,
+		Stream:        false,
+		Duration:      time.Since(startTime),
 	}, nil
 }
 
@@ -324,13 +340,14 @@ func (s *OpenAIGatewayService) handleChatStreamingResponse(
 
 	resultWithUsage := func() *OpenAIForwardResult {
 		return &OpenAIForwardResult{
-			RequestID:    requestID,
-			Usage:        usage,
-			Model:        originalModel,
-			BillingModel: mappedModel,
-			Stream:       true,
-			Duration:     time.Since(startTime),
-			FirstTokenMs: firstTokenMs,
+			RequestID:     requestID,
+			Usage:         usage,
+			Model:         originalModel,
+			BillingModel:  mappedModel,
+			UpstreamModel: mappedModel,
+			Stream:        true,
+			Duration:      time.Since(startTime),
+			FirstTokenMs:  firstTokenMs,
 		}
 	}
 

backend/internal/service/openai_gateway_messages.go

@@ -299,12 +299,13 @@ func (s *OpenAIGatewayService) handleAnthropicBufferedStreamingResponse(
 	c.JSON(http.StatusOK, anthropicResp)
 
 	return &OpenAIForwardResult{
-		RequestID:    requestID,
-		Usage:        usage,
-		Model:        originalModel,
-		BillingModel: mappedModel,
-		Stream:       false,
-		Duration:     time.Since(startTime),
+		RequestID:     requestID,
+		Usage:         usage,
+		Model:         originalModel,
+		BillingModel:  mappedModel,
+		UpstreamModel: mappedModel,
+		Stream:        false,
+		Duration:      time.Since(startTime),
 	}, nil
 }
 
@@ -347,13 +348,14 @@ func (s *OpenAIGatewayService) handleAnthropicStreamingResponse(
 	// resultWithUsage builds the final result snapshot.
 	resultWithUsage := func() *OpenAIForwardResult {
 		return &OpenAIForwardResult{
-			RequestID:    requestID,
-			Usage:        usage,
-			Model:        originalModel,
-			BillingModel: mappedModel,
-			Stream:       true,
-			Duration:     time.Since(startTime),
-			FirstTokenMs: firstTokenMs,
+			RequestID:     requestID,
+			Usage:         usage,
+			Model:         originalModel,
+			BillingModel:  mappedModel,
+			UpstreamModel: mappedModel,
+			Stream:        true,
+			Duration:      time.Since(startTime),
+			FirstTokenMs:  firstTokenMs,
 		}
 	}
 

backend/internal/service/openai_gateway_record_usage_test.go

@@ -846,7 +846,7 @@ func TestExtractOpenAIServiceTierFromBody(t *testing.T) {
 	require.Nil(t, extractOpenAIServiceTierFromBody(nil))
 }
 
-func TestOpenAIGatewayServiceRecordUsage_UsesBillingModelAndMetadataFields(t *testing.T) {
+func TestOpenAIGatewayServiceRecordUsage_UsesRequestedModelAndUpstreamModelMetadataFields(t *testing.T) {
 	usageRepo := &openAIRecordUsageLogRepoStub{inserted: true}
 	userRepo := &openAIRecordUsageUserRepoStub{}
 	subRepo := &openAIRecordUsageSubRepoStub{}
@@ -859,6 +859,7 @@ func TestOpenAIGatewayServiceRecordUsage_UsesBillingModelAndMetadataFields(t *te
 			RequestID:       "resp_billing_model_override",
 			BillingModel:    "gpt-5.1-codex",
 			Model:           "gpt-5.1",
+			UpstreamModel:   "gpt-5.1-codex",
 			ServiceTier:     &serviceTier,
 			ReasoningEffort: &reasoning,
 			Usage: OpenAIUsage{
@@ -877,7 +878,9 @@ func TestOpenAIGatewayServiceRecordUsage_UsesBillingModelAndMetadataFields(t *te
 
 	require.NoError(t, err)
 	require.NotNil(t, usageRepo.lastLog)
-	require.Equal(t, "gpt-5.1-codex", usageRepo.lastLog.Model)
+	require.Equal(t, "gpt-5.1", usageRepo.lastLog.Model)
+	require.NotNil(t, usageRepo.lastLog.UpstreamModel)
+	require.Equal(t, "gpt-5.1-codex", *usageRepo.lastLog.UpstreamModel)
 	require.NotNil(t, usageRepo.lastLog.ServiceTier)
 	require.Equal(t, serviceTier, *usageRepo.lastLog.ServiceTier)
 	require.NotNil(t, usageRepo.lastLog.ReasoningEffort)

backend/internal/service/openai_gateway_service.go

@@ -216,6 +216,9 @@ type OpenAIForwardResult struct {
 	// This is set by the Anthropic Messages conversion path where
 	// the mapped upstream model differs from the client-facing model.
 	BillingModel string
+	// UpstreamModel is the actual model sent to the upstream provider after mapping.
+	// Empty when no mapping was applied (requested model was used as-is).
+	UpstreamModel string
 	// ServiceTier records the OpenAI Responses API service tier, e.g. "priority" / "flex".
 	// Nil means the request did not specify a recognized tier.
 	ServiceTier *string
@@ -2128,6 +2131,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 				firstTokenMs,
 				wsAttempts,
 			)
+			wsResult.UpstreamModel = mappedModel
 			return wsResult, nil
 		}
 		s.writeOpenAIWSFallbackErrorResponse(c, account, wsErr)
@@ -2263,6 +2267,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 			RequestID:       resp.Header.Get("x-request-id"),
 			Usage:           *usage,
 			Model:           originalModel,
+			UpstreamModel:   mappedModel,
 			ServiceTier:     serviceTier,
 			ReasoningEffort: reasoningEffort,
 			Stream:          reqStream,
@@ -4134,7 +4139,8 @@ func (s *OpenAIGatewayService) RecordUsage(ctx context.Context, input *OpenAIRec
 		APIKeyID:              apiKey.ID,
 		AccountID:             account.ID,
 		RequestID:             requestID,
-		Model:                 billingModel,
+		Model:                 result.Model,
+		UpstreamModel:         optionalNonEqualStringPtr(result.UpstreamModel, result.Model),
 		ServiceTier:           result.ServiceTier,
 		ReasoningEffort:       result.ReasoningEffort,
 		InboundEndpoint:       optionalTrimmedStringPtr(input.InboundEndpoint),
@@ -4700,11 +4706,3 @@ func normalizeOpenAIReasoningEffort(raw string) string {
 		return ""
 	}
 }
-
-func optionalTrimmedStringPtr(raw string) *string {
-	trimmed := strings.TrimSpace(raw)
-	if trimmed == "" {
-		return nil
-	}
-	return &trimmed
-}

backend/internal/service/ops_upstream_context.go

@@ -53,6 +53,13 @@ func SetOpsLatencyMs(c *gin.Context, key string, value int64) {
 	c.Set(key, value)
 }
 
+// SetOpsUpstreamError is the exported wrapper for setOpsUpstreamError, used by
+// handler-layer code (e.g. failover-exhausted paths) that needs to record the
+// original upstream status code before mapping it to a client-facing code.
+func SetOpsUpstreamError(c *gin.Context, upstreamStatusCode int, upstreamMessage, upstreamDetail string) {
+	setOpsUpstreamError(c, upstreamStatusCode, upstreamMessage, upstreamDetail)
+}
+
 func setOpsUpstreamError(c *gin.Context, upstreamStatusCode int, upstreamMessage, upstreamDetail string) {
 	if c == nil {
 		return

backend/internal/service/overload_cooldown_test.go

@@ -0,0 +1,298 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+// ---------------------------------------------------------------------------
+// errSettingRepo: a SettingRepository that always returns errors on read
+// ---------------------------------------------------------------------------
+
+type errSettingRepo struct {
+	mockSettingRepo // embed the existing mock from backup_service_test.go
+	readErr         error
+}
+
+func (r *errSettingRepo) GetValue(_ context.Context, _ string) (string, error) {
+	return "", r.readErr
+}
+
+func (r *errSettingRepo) Get(_ context.Context, _ string) (*Setting, error) {
+	return nil, r.readErr
+}
+
+// ---------------------------------------------------------------------------
+// overloadAccountRepoStub: records SetOverloaded calls
+// ---------------------------------------------------------------------------
+
+type overloadAccountRepoStub struct {
+	mockAccountRepoForGemini
+	overloadCalls   int
+	lastOverloadID  int64
+	lastOverloadEnd time.Time
+}
+
+func (r *overloadAccountRepoStub) SetOverloaded(_ context.Context, id int64, until time.Time) error {
+	r.overloadCalls++
+	r.lastOverloadID = id
+	r.lastOverloadEnd = until
+	return nil
+}
+
+// ===========================================================================
+// SettingService: GetOverloadCooldownSettings
+// ===========================================================================
+
+func TestGetOverloadCooldownSettings_DefaultsWhenNotSet(t *testing.T) {
+	repo := newMockSettingRepo()
+	svc := NewSettingService(repo, &config.Config{})
+
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.True(t, settings.Enabled)
+	require.Equal(t, 10, settings.CooldownMinutes)
+}
+
+func TestGetOverloadCooldownSettings_ReadsFromDB(t *testing.T) {
+	repo := newMockSettingRepo()
+	data, _ := json.Marshal(OverloadCooldownSettings{Enabled: false, CooldownMinutes: 30})
+	repo.data[SettingKeyOverloadCooldownSettings] = string(data)
+	svc := NewSettingService(repo, &config.Config{})
+
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.False(t, settings.Enabled)
+	require.Equal(t, 30, settings.CooldownMinutes)
+}
+
+func TestGetOverloadCooldownSettings_ClampsMinValue(t *testing.T) {
+	repo := newMockSettingRepo()
+	data, _ := json.Marshal(OverloadCooldownSettings{Enabled: true, CooldownMinutes: 0})
+	repo.data[SettingKeyOverloadCooldownSettings] = string(data)
+	svc := NewSettingService(repo, &config.Config{})
+
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.Equal(t, 1, settings.CooldownMinutes)
+}
+
+func TestGetOverloadCooldownSettings_ClampsMaxValue(t *testing.T) {
+	repo := newMockSettingRepo()
+	data, _ := json.Marshal(OverloadCooldownSettings{Enabled: true, CooldownMinutes: 999})
+	repo.data[SettingKeyOverloadCooldownSettings] = string(data)
+	svc := NewSettingService(repo, &config.Config{})
+
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.Equal(t, 120, settings.CooldownMinutes)
+}
+
+func TestGetOverloadCooldownSettings_InvalidJSON_ReturnsDefaults(t *testing.T) {
+	repo := newMockSettingRepo()
+	repo.data[SettingKeyOverloadCooldownSettings] = "not-json"
+	svc := NewSettingService(repo, &config.Config{})
+
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.True(t, settings.Enabled)
+	require.Equal(t, 10, settings.CooldownMinutes)
+}
+
+func TestGetOverloadCooldownSettings_EmptyValue_ReturnsDefaults(t *testing.T) {
+	repo := newMockSettingRepo()
+	repo.data[SettingKeyOverloadCooldownSettings] = ""
+	svc := NewSettingService(repo, &config.Config{})
+
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.True(t, settings.Enabled)
+	require.Equal(t, 10, settings.CooldownMinutes)
+}
+
+// ===========================================================================
+// SettingService: SetOverloadCooldownSettings
+// ===========================================================================
+
+func TestSetOverloadCooldownSettings_Success(t *testing.T) {
+	repo := newMockSettingRepo()
+	svc := NewSettingService(repo, &config.Config{})
+
+	err := svc.SetOverloadCooldownSettings(context.Background(), &OverloadCooldownSettings{
+		Enabled:         false,
+		CooldownMinutes: 25,
+	})
+	require.NoError(t, err)
+
+	// Verify round-trip
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.False(t, settings.Enabled)
+	require.Equal(t, 25, settings.CooldownMinutes)
+}
+
+func TestSetOverloadCooldownSettings_RejectsNil(t *testing.T) {
+	svc := NewSettingService(newMockSettingRepo(), &config.Config{})
+	err := svc.SetOverloadCooldownSettings(context.Background(), nil)
+	require.Error(t, err)
+}
+
+func TestSetOverloadCooldownSettings_EnabledRejectsOutOfRange(t *testing.T) {
+	svc := NewSettingService(newMockSettingRepo(), &config.Config{})
+
+	for _, minutes := range []int{0, -1, 121, 999} {
+		err := svc.SetOverloadCooldownSettings(context.Background(), &OverloadCooldownSettings{
+			Enabled: true, CooldownMinutes: minutes,
+		})
+		require.Error(t, err, "should reject enabled=true + cooldown_minutes=%d", minutes)
+		require.Contains(t, err.Error(), "cooldown_minutes must be between 1-120")
+	}
+}
+
+func TestSetOverloadCooldownSettings_DisabledNormalizesOutOfRange(t *testing.T) {
+	repo := newMockSettingRepo()
+	svc := NewSettingService(repo, &config.Config{})
+
+	// enabled=false + cooldown_minutes=0 应该保存成功，值被归一化为10
+	err := svc.SetOverloadCooldownSettings(context.Background(), &OverloadCooldownSettings{
+		Enabled: false, CooldownMinutes: 0,
+	})
+	require.NoError(t, err, "disabled with invalid minutes should NOT be rejected")
+
+	// 验证持久化后读回来的值
+	settings, err := svc.GetOverloadCooldownSettings(context.Background())
+	require.NoError(t, err)
+	require.False(t, settings.Enabled)
+	require.Equal(t, 10, settings.CooldownMinutes, "should be normalized to default")
+}
+
+func TestSetOverloadCooldownSettings_AcceptsBoundaries(t *testing.T) {
+	svc := NewSettingService(newMockSettingRepo(), &config.Config{})
+
+	for _, minutes := range []int{1, 60, 120} {
+		err := svc.SetOverloadCooldownSettings(context.Background(), &OverloadCooldownSettings{
+			Enabled: true, CooldownMinutes: minutes,
+		})
+		require.NoError(t, err, "should accept cooldown_minutes=%d", minutes)
+	}
+}
+
+// ===========================================================================
+// RateLimitService: handle529 behaviour
+// ===========================================================================
+
+func TestHandle529_EnabledFromDB_PausesAccount(t *testing.T) {
+	accountRepo := &overloadAccountRepoStub{}
+	settingRepo := newMockSettingRepo()
+	data, _ := json.Marshal(OverloadCooldownSettings{Enabled: true, CooldownMinutes: 15})
+	settingRepo.data[SettingKeyOverloadCooldownSettings] = string(data)
+
+	settingSvc := NewSettingService(settingRepo, &config.Config{})
+	svc := NewRateLimitService(accountRepo, nil, &config.Config{}, nil, nil)
+	svc.SetSettingService(settingSvc)
+
+	account := &Account{ID: 42, Platform: PlatformAnthropic, Type: AccountTypeOAuth}
+	before := time.Now()
+	svc.handle529(context.Background(), account)
+
+	require.Equal(t, 1, accountRepo.overloadCalls)
+	require.Equal(t, int64(42), accountRepo.lastOverloadID)
+	require.WithinDuration(t, before.Add(15*time.Minute), accountRepo.lastOverloadEnd, 2*time.Second)
+}
+
+func TestHandle529_DisabledFromDB_SkipsAccount(t *testing.T) {
+	accountRepo := &overloadAccountRepoStub{}
+	settingRepo := newMockSettingRepo()
+	data, _ := json.Marshal(OverloadCooldownSettings{Enabled: false, CooldownMinutes: 15})
+	settingRepo.data[SettingKeyOverloadCooldownSettings] = string(data)
+
+	settingSvc := NewSettingService(settingRepo, &config.Config{})
+	svc := NewRateLimitService(accountRepo, nil, &config.Config{}, nil, nil)
+	svc.SetSettingService(settingSvc)
+
+	account := &Account{ID: 42, Platform: PlatformAnthropic, Type: AccountTypeOAuth}
+	svc.handle529(context.Background(), account)
+
+	require.Equal(t, 0, accountRepo.overloadCalls, "should NOT pause when disabled")
+}
+
+func TestHandle529_NilSettingService_FallsBackToConfig(t *testing.T) {
+	accountRepo := &overloadAccountRepoStub{}
+	cfg := &config.Config{}
+	cfg.RateLimit.OverloadCooldownMinutes = 20
+	svc := NewRateLimitService(accountRepo, nil, cfg, nil, nil)
+	// NOT calling SetSettingService — remains nil
+
+	account := &Account{ID: 77, Platform: PlatformAnthropic, Type: AccountTypeOAuth}
+	before := time.Now()
+	svc.handle529(context.Background(), account)
+
+	require.Equal(t, 1, accountRepo.overloadCalls)
+	require.WithinDuration(t, before.Add(20*time.Minute), accountRepo.lastOverloadEnd, 2*time.Second)
+}
+
+func TestHandle529_NilSettingService_ZeroConfig_DefaultsTen(t *testing.T) {
+	accountRepo := &overloadAccountRepoStub{}
+	svc := NewRateLimitService(accountRepo, nil, &config.Config{}, nil, nil)
+
+	account := &Account{ID: 88, Platform: PlatformAnthropic, Type: AccountTypeOAuth}
+	before := time.Now()
+	svc.handle529(context.Background(), account)
+
+	require.Equal(t, 1, accountRepo.overloadCalls)
+	require.WithinDuration(t, before.Add(10*time.Minute), accountRepo.lastOverloadEnd, 2*time.Second)
+}
+
+func TestHandle529_DBReadError_FallsBackToConfig(t *testing.T) {
+	accountRepo := &overloadAccountRepoStub{}
+	errRepo := &errSettingRepo{readErr: context.DeadlineExceeded}
+	errRepo.data = make(map[string]string)
+
+	cfg := &config.Config{}
+	cfg.RateLimit.OverloadCooldownMinutes = 7
+	settingSvc := NewSettingService(errRepo, cfg)
+	svc := NewRateLimitService(accountRepo, nil, cfg, nil, nil)
+	svc.SetSettingService(settingSvc)
+
+	account := &Account{ID: 99, Platform: PlatformAnthropic, Type: AccountTypeOAuth}
+	before := time.Now()
+	svc.handle529(context.Background(), account)
+
+	require.Equal(t, 1, accountRepo.overloadCalls)
+	require.WithinDuration(t, before.Add(7*time.Minute), accountRepo.lastOverloadEnd, 2*time.Second)
+}
+
+// ===========================================================================
+// Model: defaults & JSON round-trip
+// ===========================================================================
+
+func TestDefaultOverloadCooldownSettings(t *testing.T) {
+	d := DefaultOverloadCooldownSettings()
+	require.True(t, d.Enabled)
+	require.Equal(t, 10, d.CooldownMinutes)
+}
+
+func TestOverloadCooldownSettings_JSONRoundTrip(t *testing.T) {
+	original := OverloadCooldownSettings{Enabled: false, CooldownMinutes: 42}
+	data, err := json.Marshal(original)
+	require.NoError(t, err)
+
+	var decoded OverloadCooldownSettings
+	require.NoError(t, json.Unmarshal(data, &decoded))
+	require.Equal(t, original, decoded)
+
+	// Verify JSON uses snake_case field names
+	var raw map[string]any
+	require.NoError(t, json.Unmarshal(data, &raw))
+	_, hasEnabled := raw["enabled"]
+	_, hasCooldown := raw["cooldown_minutes"]
+	require.True(t, hasEnabled, "JSON must use 'enabled'")
+	require.True(t, hasCooldown, "JSON must use 'cooldown_minutes'")
+}

backend/internal/service/ratelimit_service.go

@@ -1023,11 +1023,34 @@ func parseOpenAIRateLimitResetTime(body []byte) *int64 {
 }
 
 // handle529 处理529过载错误
-// 根据配置设置过载冷却时间
+// 根据配置决定是否暂停账号调度及冷却时长
 func (s *RateLimitService) handle529(ctx context.Context, account *Account) {
-	cooldownMinutes := s.cfg.RateLimit.OverloadCooldownMinutes
+	var settings *OverloadCooldownSettings
+	if s.settingService != nil {
+		var err error
+		settings, err = s.settingService.GetOverloadCooldownSettings(ctx)
+		if err != nil {
+			slog.Warn("overload_settings_read_failed", "account_id", account.ID, "error", err)
+			settings = nil
+		}
+	}
+	// 回退到配置文件
+	if settings == nil {
+		cooldown := s.cfg.RateLimit.OverloadCooldownMinutes
+		if cooldown <= 0 {
+			cooldown = 10
+		}
+		settings = &OverloadCooldownSettings{Enabled: true, CooldownMinutes: cooldown}
+	}
+
+	if !settings.Enabled {
+		slog.Info("account_529_ignored", "account_id", account.ID, "reason", "overload_cooldown_disabled")
+		return
+	}
+
+	cooldownMinutes := settings.CooldownMinutes
 	if cooldownMinutes <= 0 {
-		cooldownMinutes = 10 // 默认10分钟
+		cooldownMinutes = 10
 	}
 
 	until := time.Now().Add(time.Duration(cooldownMinutes) * time.Minute)
@@ -1087,10 +1110,13 @@ func (s *RateLimitService) UpdateSessionWindow(ctx context.Context, account *Acc
 		slog.Info("account_session_window_initialized", "account_id", account.ID, "window_start", start, "window_end", end, "status", status)
 	}
 
-	// 窗口重置时清除旧的 utilization，避免残留上个窗口的数据
+	// 窗口重置时清除旧的 utilization 和被动采样数据，避免残留上个窗口的数据
 	if windowEnd != nil && needInitWindow {
 		_ = s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
-			"session_window_utilization": nil,
+			"session_window_utilization":   nil,
+			"passive_usage_7d_utilization": nil,
+			"passive_usage_7d_reset":       nil,
+			"passive_usage_sampled_at":     nil,
 		})
 	}
 
@@ -1098,14 +1124,33 @@ func (s *RateLimitService) UpdateSessionWindow(ctx context.Context, account *Acc
 		slog.Warn("session_window_update_failed", "account_id", account.ID, "error", err)
 	}
 
-	// 存储真实的 utilization 值（0-1 小数），供 estimateSetupTokenUsage 使用
+	// 被动采样：从响应头收集 5h + 7d utilization，合并为一次 DB 写入
+	extraUpdates := make(map[string]any, 4)
+	// 5h utilization（0-1 小数），供 estimateSetupTokenUsage 使用
 	if utilStr := headers.Get("anthropic-ratelimit-unified-5h-utilization"); utilStr != "" {
 		if util, err := strconv.ParseFloat(utilStr, 64); err == nil {
-			if err := s.accountRepo.UpdateExtra(ctx, account.ID, map[string]any{
-				"session_window_utilization": util,
-			}); err != nil {
-				slog.Warn("session_window_utilization_update_failed", "account_id", account.ID, "error", err)
+			extraUpdates["session_window_utilization"] = util
+		}
+	}
+	// 7d utilization（0-1 小数）
+	if utilStr := headers.Get("anthropic-ratelimit-unified-7d-utilization"); utilStr != "" {
+		if util, err := strconv.ParseFloat(utilStr, 64); err == nil {
+			extraUpdates["passive_usage_7d_utilization"] = util
+		}
+	}
+	// 7d reset timestamp
+	if resetStr := headers.Get("anthropic-ratelimit-unified-7d-reset"); resetStr != "" {
+		if ts, err := strconv.ParseInt(resetStr, 10, 64); err == nil {
+			if ts > 1e11 {
+				ts = ts / 1000
 			}
+			extraUpdates["passive_usage_7d_reset"] = ts
+		}
+	}
+	if len(extraUpdates) > 0 {
+		extraUpdates["passive_usage_sampled_at"] = time.Now().UTC().Format(time.RFC3339)
+		if err := s.accountRepo.UpdateExtra(ctx, account.ID, extraUpdates); err != nil {
+			slog.Warn("passive_usage_update_failed", "account_id", account.ID, "error", err)
 		}
 	}
 

backend/internal/service/setting_service.go

@@ -1172,6 +1172,57 @@ func (s *SettingService) GetLinuxDoConnectOAuthConfig(ctx context.Context) (conf
 	return effective, nil
 }
 
+// GetOverloadCooldownSettings 获取529过载冷却配置
+func (s *SettingService) GetOverloadCooldownSettings(ctx context.Context) (*OverloadCooldownSettings, error) {
+	value, err := s.settingRepo.GetValue(ctx, SettingKeyOverloadCooldownSettings)
+	if err != nil {
+		if errors.Is(err, ErrSettingNotFound) {
+			return DefaultOverloadCooldownSettings(), nil
+		}
+		return nil, fmt.Errorf("get overload cooldown settings: %w", err)
+	}
+	if value == "" {
+		return DefaultOverloadCooldownSettings(), nil
+	}
+
+	var settings OverloadCooldownSettings
+	if err := json.Unmarshal([]byte(value), &settings); err != nil {
+		return DefaultOverloadCooldownSettings(), nil
+	}
+
+	// 修正配置值范围
+	if settings.CooldownMinutes < 1 {
+		settings.CooldownMinutes = 1
+	}
+	if settings.CooldownMinutes > 120 {
+		settings.CooldownMinutes = 120
+	}
+
+	return &settings, nil
+}
+
+// SetOverloadCooldownSettings 设置529过载冷却配置
+func (s *SettingService) SetOverloadCooldownSettings(ctx context.Context, settings *OverloadCooldownSettings) error {
+	if settings == nil {
+		return fmt.Errorf("settings cannot be nil")
+	}
+
+	// 禁用时修正为合法值即可，不拒绝请求
+	if settings.CooldownMinutes < 1 || settings.CooldownMinutes > 120 {
+		if settings.Enabled {
+			return fmt.Errorf("cooldown_minutes must be between 1-120")
+		}
+		settings.CooldownMinutes = 10 // 禁用状态下归一化为默认值
+	}
+
+	data, err := json.Marshal(settings)
+	if err != nil {
+		return fmt.Errorf("marshal overload cooldown settings: %w", err)
+	}
+
+	return s.settingRepo.Set(ctx, SettingKeyOverloadCooldownSettings, string(data))
+}
+
 // GetStreamTimeoutSettings 获取流超时处理配置
 func (s *SettingService) GetStreamTimeoutSettings(ctx context.Context) (*StreamTimeoutSettings, error) {
 	value, err := s.settingRepo.GetValue(ctx, SettingKeyStreamTimeoutSettings)

backend/internal/service/settings_view.go

@@ -222,6 +222,22 @@ type BetaPolicySettings struct {
 	Rules []BetaPolicyRule `json:"rules"`
 }
 
+// OverloadCooldownSettings 529过载冷却配置
+type OverloadCooldownSettings struct {
+	// Enabled 是否在收到529时暂停账号调度
+	Enabled bool `json:"enabled"`
+	// CooldownMinutes 冷却时长（分钟）
+	CooldownMinutes int `json:"cooldown_minutes"`
+}
+
+// DefaultOverloadCooldownSettings 返回默认的过载冷却配置（启用，10分钟）
+func DefaultOverloadCooldownSettings() *OverloadCooldownSettings {
+	return &OverloadCooldownSettings{
+		Enabled:         true,
+		CooldownMinutes: 10,
+	}
+}
+
 // DefaultBetaPolicySettings 返回默认的 Beta 策略配置
 func DefaultBetaPolicySettings() *BetaPolicySettings {
 	return &BetaPolicySettings{

backend/internal/service/usage_log.go

@@ -98,6 +98,9 @@ type UsageLog struct {
 	AccountID int64
 	RequestID string
 	Model     string
+	// UpstreamModel is the actual model sent to the upstream provider after mapping.
+	// Nil means no mapping was applied (requested model was used as-is).
+	UpstreamModel *string
 	// ServiceTier records the OpenAI service tier used for billing, e.g. "priority" / "flex".
 	ServiceTier *string
 	// ReasoningEffort is the request's reasoning effort level.

backend/internal/service/usage_log_helpers.go

@@ -0,0 +1,21 @@
+package service
+
+import "strings"
+
+func optionalTrimmedStringPtr(raw string) *string {
+	trimmed := strings.TrimSpace(raw)
+	if trimmed == "" {
+		return nil
+	}
+	return &trimmed
+}
+
+// optionalNonEqualStringPtr returns a pointer to value if it is non-empty and
+// differs from compare; otherwise nil. Used to store upstream_model only when
+// it differs from the requested model.
+func optionalNonEqualStringPtr(value, compare string) *string {
+	if value == "" || value == compare {
+		return nil
+	}
+	return &value
+}

backend/internal/setup/handler.go

@@ -247,6 +247,12 @@ func install(c *gin.Context) {
 		return
 	}
 
+	req.Admin.Email = strings.TrimSpace(req.Admin.Email)
+	req.Database.Host = strings.TrimSpace(req.Database.Host)
+	req.Database.User = strings.TrimSpace(req.Database.User)
+	req.Database.DBName = strings.TrimSpace(req.Database.DBName)
+	req.Redis.Host = strings.TrimSpace(req.Redis.Host)
+
 	// ========== COMPREHENSIVE INPUT VALIDATION ==========
 	// Database validation
 	if !validateHostname(req.Database.Host) {
@@ -319,13 +325,6 @@ func install(c *gin.Context) {
 		return
 	}
 
-	// Trim whitespace from string inputs
-	req.Admin.Email = strings.TrimSpace(req.Admin.Email)
-	req.Database.Host = strings.TrimSpace(req.Database.Host)
-	req.Database.User = strings.TrimSpace(req.Database.User)
-	req.Database.DBName = strings.TrimSpace(req.Database.DBName)
-	req.Redis.Host = strings.TrimSpace(req.Redis.Host)
-
 	cfg := &SetupConfig{
 		Database: req.Database,
 		Redis:    req.Redis,

backend/internal/web/embed_on.go

@@ -180,7 +180,37 @@ func (s *FrontendServer) injectSettings(settingsJSON []byte) []byte {
 
 	// Inject before </head>
 	headClose := []byte("</head>")
-	return bytes.Replace(s.baseHTML, headClose, append(script, headClose...), 1)
+	result := bytes.Replace(s.baseHTML, headClose, append(script, headClose...), 1)
+
+	// Replace <title> with custom site name so the browser tab shows it immediately
+	result = injectSiteTitle(result, settingsJSON)
+
+	return result
+}
+
+// injectSiteTitle replaces the static <title> in HTML with the configured site name.
+// This ensures the browser tab shows the correct title before JS executes.
+func injectSiteTitle(html, settingsJSON []byte) []byte {
+	var cfg struct {
+		SiteName string `json:"site_name"`
+	}
+	if err := json.Unmarshal(settingsJSON, &cfg); err != nil || cfg.SiteName == "" {
+		return html
+	}
+
+	// Find and replace the existing <title>...</title>
+	titleStart := bytes.Index(html, []byte("<title>"))
+	titleEnd := bytes.Index(html, []byte("</title>"))
+	if titleStart == -1 || titleEnd == -1 || titleEnd <= titleStart {
+		return html
+	}
+
+	newTitle := []byte("<title>" + cfg.SiteName + " - AI API Gateway</title>")
+	var buf bytes.Buffer
+	buf.Write(html[:titleStart])
+	buf.Write(newTitle)
+	buf.Write(html[titleEnd+len("</title>"):])
+	return buf.Bytes()
 }
 
 // replaceNoncePlaceholder replaces the nonce placeholder with actual nonce value

backend/internal/web/embed_test.go

@@ -20,6 +20,78 @@ func init() {
 	gin.SetMode(gin.TestMode)
 }
 
+func TestInjectSiteTitle(t *testing.T) {
+	t.Run("replaces_title_with_site_name", func(t *testing.T) {
+		html := []byte(`<html><head><title>Sub2API - AI API Gateway</title></head><body></body></html>`)
+		settingsJSON := []byte(`{"site_name":"MyCustomSite"}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		assert.Contains(t, string(result), "<title>MyCustomSite - AI API Gateway</title>")
+		assert.NotContains(t, string(result), "Sub2API")
+	})
+
+	t.Run("returns_unchanged_when_site_name_empty", func(t *testing.T) {
+		html := []byte(`<html><head><title>Sub2API - AI API Gateway</title></head><body></body></html>`)
+		settingsJSON := []byte(`{"site_name":""}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		assert.Equal(t, string(html), string(result))
+	})
+
+	t.Run("returns_unchanged_when_site_name_missing", func(t *testing.T) {
+		html := []byte(`<html><head><title>Sub2API - AI API Gateway</title></head><body></body></html>`)
+		settingsJSON := []byte(`{"other_field":"value"}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		assert.Equal(t, string(html), string(result))
+	})
+
+	t.Run("returns_unchanged_when_invalid_json", func(t *testing.T) {
+		html := []byte(`<html><head><title>Sub2API - AI API Gateway</title></head><body></body></html>`)
+		settingsJSON := []byte(`{invalid json}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		assert.Equal(t, string(html), string(result))
+	})
+
+	t.Run("returns_unchanged_when_no_title_tag", func(t *testing.T) {
+		html := []byte(`<html><head></head><body></body></html>`)
+		settingsJSON := []byte(`{"site_name":"MyCustomSite"}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		assert.Equal(t, string(html), string(result))
+	})
+
+	t.Run("returns_unchanged_when_title_has_attributes", func(t *testing.T) {
+		// The function looks for "<title>" literally, so attributes are not supported
+		// This is acceptable since index.html uses plain <title> without attributes
+		html := []byte(`<html><head><title lang="en">Sub2API</title></head><body></body></html>`)
+		settingsJSON := []byte(`{"site_name":"NewSite"}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		// Should return unchanged since <title> with attributes is not matched
+		assert.Equal(t, string(html), string(result))
+	})
+
+	t.Run("preserves_rest_of_html", func(t *testing.T) {
+		html := []byte(`<html><head><meta charset="UTF-8"><title>Sub2API</title><script src="app.js"></script></head><body><div id="app"></div></body></html>`)
+		settingsJSON := []byte(`{"site_name":"TestSite"}`)
+
+		result := injectSiteTitle(html, settingsJSON)
+
+		assert.Contains(t, string(result), `<meta charset="UTF-8">`)
+		assert.Contains(t, string(result), `<script src="app.js"></script>`)
+		assert.Contains(t, string(result), `<div id="app"></div>`)
+		assert.Contains(t, string(result), "<title>TestSite - AI API Gateway</title>")
+	})
+}
+
 func TestReplaceNoncePlaceholder(t *testing.T) {
 	t.Run("replaces_single_placeholder", func(t *testing.T) {
 		html := []byte(`<script nonce="__CSP_NONCE_VALUE__">console.log('test');</script>`)

backend/migrations/075_add_usage_log_upstream_model.sql

@@ -0,0 +1,4 @@
+-- Add upstream_model field to usage_logs.
+-- Stores the actual upstream model name when it differs from the requested model
+-- (i.e., when model mapping is applied). NULL means no mapping was applied.
+ALTER TABLE usage_logs ADD COLUMN IF NOT EXISTS upstream_model VARCHAR(100);

backend/migrations/075_map_haiku45_to_sonnet46.sql

@@ -0,0 +1,17 @@
+-- Map claude-haiku-4-5 variants target from claude-sonnet-4-5 to claude-sonnet-4-6
+--
+-- Only updates when the current target is exactly claude-sonnet-4-5.
+
+-- 1. claude-haiku-4-5
+UPDATE accounts
+SET credentials = jsonb_set(credentials, '{model_mapping,claude-haiku-4-5}', '"claude-sonnet-4-6"')
+WHERE platform = 'antigravity'
+  AND deleted_at IS NULL
+  AND credentials->'model_mapping'->>'claude-haiku-4-5' = 'claude-sonnet-4-5';
+
+-- 2. claude-haiku-4-5-20251001
+UPDATE accounts
+SET credentials = jsonb_set(credentials, '{model_mapping,claude-haiku-4-5-20251001}', '"claude-sonnet-4-6"')
+WHERE platform = 'antigravity'
+  AND deleted_at IS NULL
+  AND credentials->'model_mapping'->>'claude-haiku-4-5-20251001' = 'claude-sonnet-4-5';

backend/migrations/076_add_usage_log_upstream_model_index_notx.sql

@@ -0,0 +1,3 @@
+-- Support upstream_model / mapping model distribution aggregations with time-range filters.
+CREATE INDEX CONCURRENTLY IF NOT EXISTS idx_usage_logs_created_model_upstream_model
+ON usage_logs (created_at, model, upstream_model);

backend/migrations/README.md

@@ -34,18 +34,18 @@ Example: `017_add_gemini_tier_id.sql`
 
 ## Migration File Structure
 
-```sql
--- +goose Up
--- +goose StatementBegin
--- Your forward migration SQL here
--- +goose StatementEnd
+This project uses a custom migration runner (`internal/repository/migrations_runner.go`) that executes the full SQL file content as-is.
 
--- +goose Down
--- +goose StatementBegin
--- Your rollback migration SQL here
--- +goose StatementEnd
+- Regular migrations (`*.sql`): executed in a transaction.
+- Non-transactional migrations (`*_notx.sql`): split by statement and executed without transaction (for `CONCURRENTLY`).
+
+```sql
+-- Forward-only migration (recommended)
+ALTER TABLE usage_logs ADD COLUMN IF NOT EXISTS example_column VARCHAR(100);
 ```
 
+> ⚠️ Do **not** place executable "Down" SQL in the same file. The runner does not parse goose Up/Down sections and will execute all SQL statements in the file.
+
 ## Important Rules
 
 ### ⚠️ Immutability Principle
@@ -66,9 +66,9 @@ Why?
    touch migrations/018_your_change.sql
    ```
 
-2. **Write Up and Down migrations**
-   - Up: Apply the change
-   - Down: Revert the change (should be symmetric with Up)
+2. **Write forward-only migration SQL**
+   - Put only the intended schema change in the file
+   - If rollback is needed, create a new migration file to revert
 
 3. **Test locally**
    ```bash
@@ -144,8 +144,6 @@ touch migrations/018_your_new_change.sql
 ## Example Migration
 
 ```sql
--- +goose Up
--- +goose StatementBegin
 -- Add tier_id field to Gemini OAuth accounts for quota tracking
 UPDATE accounts
 SET credentials = jsonb_set(
@@ -157,17 +155,6 @@ SET credentials = jsonb_set(
 WHERE platform = 'gemini'
   AND type = 'oauth'
   AND credentials->>'tier_id' IS NULL;
--- +goose StatementEnd
-
--- +goose Down
--- +goose StatementBegin
--- Remove tier_id field
-UPDATE accounts
-SET credentials = credentials - 'tier_id'
-WHERE platform = 'gemini'
-  AND type = 'oauth'
-  AND credentials->>'tier_id' = 'LEGACY';
--- +goose StatementEnd
 ```
 
 ## Troubleshooting
@@ -194,5 +181,4 @@ VALUES ('NNN_migration.sql', 'calculated_checksum', NOW());
 ## References
 
 - Migration runner: `internal/repository/migrations_runner.go`
-- Goose syntax: https://github.com/pressly/goose
 - PostgreSQL docs: https://www.postgresql.org/docs/

deploy/docker-compose.local.yml

@@ -38,7 +38,7 @@ services:
       - ./data:/app/data
       # Optional: Mount custom config.yaml (uncomment and create the file first)
       # Copy config.example.yaml to config.yaml, modify it, then uncomment:
-      # - ./config.yaml:/app/data/config.yaml:ro
+      # - ./config.yaml:/app/data/config.yaml
     environment:
       # =======================================================================
       # Auto Setup (REQUIRED for Docker deployment)

deploy/docker-compose.yml

@@ -30,7 +30,7 @@ services:
       - sub2api_data:/app/data
       # Optional: Mount custom config.yaml (uncomment and create the file first)
       # Copy config.example.yaml to config.yaml, modify it, then uncomment:
-      # - ./config.yaml:/app/data/config.yaml:ro
+      # - ./config.yaml:/app/data/config.yaml
     environment:
       # =======================================================================
       # Auto Setup (REQUIRED for Docker deployment)

deploy/docker-entrypoint.sh

@@ -6,7 +6,8 @@ set -e
 # preventing the non-root sub2api user from writing files.
 if [ "$(id -u)" = "0" ]; then
     mkdir -p /app/data
-    chown -R sub2api:sub2api /app/data
+    # Use || true to avoid failure on read-only mounted files (e.g. config.yaml:ro)
+    chown -R sub2api:sub2api /app/data 2>/dev/null || true
     # Re-invoke this script as sub2api so the flag-detection below
     # also runs under the correct user.
     exec su-exec sub2api "$0" "$@"

frontend/package.json

@@ -16,6 +16,7 @@
   },
   "dependencies": {
     "@lobehub/icons": "^4.0.2",
+    "@tanstack/vue-virtual": "^3.13.23",
     "@vueuse/core": "^10.7.0",
     "axios": "^1.13.5",
     "chart.js": "^4.4.1",

frontend/pnpm-lock.yaml

@@ -11,6 +11,9 @@ importers:
       '@lobehub/icons':
         specifier: ^4.0.2
         version: 4.0.2(@lobehub/ui@4.9.2)(@types/react@19.2.7)(antd@6.1.3(react-dom@19.2.3(react@19.2.3))(react@19.2.3))(react-dom@19.2.3(react@19.2.3))(react@19.2.3)
+      '@tanstack/vue-virtual':
+        specifier: ^3.13.23
+        version: 3.13.23(vue@3.5.26(typescript@5.6.3))
       '@vueuse/core':
         specifier: ^10.7.0
         version: 10.11.1(vue@3.5.26(typescript@5.6.3))
@@ -1376,6 +1379,14 @@ packages:
     peerDependencies:
       react: '>= 16.3.0'
 
+  '@tanstack/virtual-core@3.13.23':
+    resolution: {integrity: sha512-zSz2Z2HNyLjCplANTDyl3BcdQJc2k1+yyFoKhNRmCr7V7dY8o8q5m8uFTI1/Pg1kL+Hgrz6u3Xo6eFUB7l66cg==}
+
+  '@tanstack/vue-virtual@3.13.23':
+    resolution: {integrity: sha512-b5jPluAR6U3eOq6GWAYSpj3ugnAIZgGR0e6aGAgyRse0Yu6MVQQ0ZWm9SArSXWtageogn6bkVD8D//c4IjW3xQ==}
+    peerDependencies:
+      vue: ^2.7.0 || ^3.0.0
+
   '@types/d3-array@3.2.2':
     resolution: {integrity: sha512-hOLWVbm7uRza0BYXpIIW5pxfrKe0W+D5lrFiAEYR+pb6w3N2SwSMaJbXdUfSEv+dT4MfHBLtn5js0LAWaO6otw==}
 
@@ -5808,6 +5819,13 @@ snapshots:
     dependencies:
       react: 19.2.3
 
+  '@tanstack/virtual-core@3.13.23': {}
+
+  '@tanstack/vue-virtual@3.13.23(vue@3.5.26(typescript@5.6.3))':
+    dependencies:
+      '@tanstack/virtual-core': 3.13.23
+      vue: 3.5.26(typescript@5.6.3)
+
   '@types/d3-array@3.2.2': {}
 
   '@types/d3-axis@3.0.6':

frontend/src/App.vue

@@ -3,6 +3,7 @@ import { RouterView, useRouter, useRoute } from 'vue-router'
 import { onMounted, onBeforeUnmount, watch } from 'vue'
 import Toast from '@/components/common/Toast.vue'
 import NavigationProgress from '@/components/common/NavigationProgress.vue'
+import { resolveDocumentTitle } from '@/router/title'
 import AnnouncementPopup from '@/components/common/AnnouncementPopup.vue'
 import { useAppStore, useAuthStore, useSubscriptionStore, useAnnouncementStore } from '@/stores'
 import { getSetupStatus } from '@/api/setup'
@@ -104,6 +105,9 @@ onMounted(async () => {
 
   // Load public settings into appStore (will be cached for other components)
   await appStore.fetchPublicSettings()
+
+  // Re-resolve document title now that siteName is available
+  document.title = resolveDocumentTitle(route.meta.title, appStore.siteName, route.meta.titleKey as string)
 })
 </script>
 

frontend/src/api/admin/accounts.ts

@@ -66,6 +66,7 @@ export async function listWithEtag(
     platform?: string
     type?: string
     status?: string
+    group?: string
     search?: string
     lite?: string
   },
@@ -223,8 +224,10 @@ export async function clearError(id: number): Promise<Account> {
  * @param id - Account ID
  * @returns Account usage info
  */
-export async function getUsage(id: number): Promise<AccountUsageInfo> {
-  const { data } = await apiClient.get<AccountUsageInfo>(`/admin/accounts/${id}/usage`)
+export async function getUsage(id: number, source?: 'passive' | 'active'): Promise<AccountUsageInfo> {
+  const { data } = await apiClient.get<AccountUsageInfo>(`/admin/accounts/${id}/usage`, {
+    params: source ? { source } : undefined
+  })
   return data
 }
 

frontend/src/api/admin/dashboard.ts

@@ -81,6 +81,7 @@ export interface ModelStatsParams {
   user_id?: number
   api_key_id?: number
   model?: string
+  model_source?: 'requested' | 'upstream' | 'mapping'
   account_id?: number
   group_id?: number
   request_type?: UsageRequestType
@@ -162,6 +163,7 @@ export interface UserBreakdownParams {
   end_date?: string
   group_id?: number
   model?: string
+  model_source?: 'requested' | 'upstream' | 'mapping'
   endpoint?: string
   endpoint_type?: 'inbound' | 'upstream' | 'path'
   limit?: number

frontend/src/api/admin/settings.ts

@@ -242,6 +242,33 @@ export async function deleteAdminApiKey(): Promise<{ message: string }> {
   return data
 }
 
+// ==================== Overload Cooldown Settings ====================
+
+/**
+ * Overload cooldown settings interface (529 handling)
+ */
+export interface OverloadCooldownSettings {
+  enabled: boolean
+  cooldown_minutes: number
+}
+
+export async function getOverloadCooldownSettings(): Promise<OverloadCooldownSettings> {
+  const { data } = await apiClient.get<OverloadCooldownSettings>('/admin/settings/overload-cooldown')
+  return data
+}
+
+export async function updateOverloadCooldownSettings(
+  settings: OverloadCooldownSettings
+): Promise<OverloadCooldownSettings> {
+  const { data } = await apiClient.put<OverloadCooldownSettings>(
+    '/admin/settings/overload-cooldown',
+    settings
+  )
+  return data
+}
+
+// ==================== Stream Timeout Settings ====================
+
 /**
  * Stream timeout settings interface
  */
@@ -499,6 +526,8 @@ export const settingsAPI = {
   getAdminApiKey,
   regenerateAdminApiKey,
   deleteAdminApiKey,
+  getOverloadCooldownSettings,
+  updateOverloadCooldownSettings,
   getStreamTimeoutSettings,
   updateStreamTimeoutSettings,
   getRectifierSettings,

frontend/src/components/account/AccountUsageCell.vue

@@ -67,6 +67,38 @@
           :resets-at="usageInfo.seven_day_sonnet.resets_at"
           color="purple"
         />
+
+        <!-- Passive sampling label + active query button -->
+        <div class="flex items-center gap-1.5 mt-0.5">
+          <span
+            v-if="usageInfo.source === 'passive'"
+            class="text-[9px] text-gray-400 dark:text-gray-500 italic"
+          >
+            {{ t('admin.accounts.usageWindow.passiveSampled') }}
+          </span>
+          <button
+            type="button"
+            class="inline-flex items-center gap-0.5 rounded px-1.5 py-0.5 text-[9px] font-medium text-blue-600 hover:bg-blue-50 dark:text-blue-400 dark:hover:bg-blue-900/30 transition-colors"
+            :disabled="activeQueryLoading"
+            @click="loadActiveUsage"
+          >
+            <svg
+              class="h-2.5 w-2.5"
+              :class="{ 'animate-spin': activeQueryLoading }"
+              fill="none"
+              stroke="currentColor"
+              viewBox="0 0 24 24"
+            >
+              <path
+                stroke-linecap="round"
+                stroke-linejoin="round"
+                stroke-width="2"
+                d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15"
+              />
+            </svg>
+            {{ t('admin.accounts.usageWindow.activeQuery') }}
+          </button>
+        </div>
       </div>
 
       <!-- No data yet -->
@@ -433,6 +465,7 @@ const props = withDefaults(
 const { t } = useI18n()
 
 const loading = ref(false)
+const activeQueryLoading = ref(false)
 const error = ref<string | null>(null)
 const usageInfo = ref<AccountUsageInfo | null>(null)
 
@@ -888,14 +921,18 @@ const copyValidationURL = async () => {
   }
 }
 
-const loadUsage = async () => {
+const isAnthropicOAuthOrSetupToken = computed(() => {
+  return props.account.platform === 'anthropic' && (props.account.type === 'oauth' || props.account.type === 'setup-token')
+})
+
+const loadUsage = async (source?: 'passive' | 'active') => {
   if (!shouldFetchUsage.value) return
 
   loading.value = true
   error.value = null
 
   try {
-    usageInfo.value = await adminAPI.accounts.getUsage(props.account.id)
+    usageInfo.value = await adminAPI.accounts.getUsage(props.account.id, source)
   } catch (e: any) {
     error.value = t('common.error')
     console.error('Failed to load usage:', e)
@@ -904,6 +941,17 @@ const loadUsage = async () => {
   }
 }
 
+const loadActiveUsage = async () => {
+  activeQueryLoading.value = true
+  try {
+    usageInfo.value = await adminAPI.accounts.getUsage(props.account.id, 'active')
+  } catch (e: any) {
+    console.error('Failed to load active usage:', e)
+  } finally {
+    activeQueryLoading.value = false
+  }
+}
+
 // ===== API Key quota progress bars =====
 
 interface QuotaBarInfo {
@@ -993,7 +1041,8 @@ const formatKeyUserCost = computed(() => {
 
 onMounted(() => {
   if (!shouldAutoLoadUsageOnMount.value) return
-  loadUsage()
+  const source = isAnthropicOAuthOrSetupToken.value ? 'passive' : undefined
+  loadUsage(source)
 })
 
 watch(openAIUsageRefreshKey, (nextKey, prevKey) => {
@@ -1011,7 +1060,8 @@ watch(
     if (nextToken === prevToken) return
     if (!shouldFetchUsage.value) return
 
-    loadUsage().catch((e) => {
+    const source = isAnthropicOAuthOrSetupToken.value ? 'passive' : undefined
+    loadUsage(source).catch((e) => {
       console.error('Failed to refresh usage after manual refresh:', e)
     })
   }

frontend/src/components/admin/account/AccountTableFilters.vue

@@ -26,5 +26,9 @@ const updateGroup = (value: string | number | boolean | null) => { emit('update:
 const pOpts = computed(() => [{ value: '', label: t('admin.accounts.allPlatforms') }, { value: 'anthropic', label: 'Anthropic' }, { value: 'openai', label: 'OpenAI' }, { value: 'gemini', label: 'Gemini' }, { value: 'antigravity', label: 'Antigravity' }, { value: 'sora', label: 'Sora' }])
 const tOpts = computed(() => [{ value: '', label: t('admin.accounts.allTypes') }, { value: 'oauth', label: t('admin.accounts.oauthType') }, { value: 'setup-token', label: t('admin.accounts.setupToken') }, { value: 'apikey', label: t('admin.accounts.apiKey') }, { value: 'bedrock', label: 'AWS Bedrock' }])
 const sOpts = computed(() => [{ value: '', label: t('admin.accounts.allStatus') }, { value: 'active', label: t('admin.accounts.status.active') }, { value: 'inactive', label: t('admin.accounts.status.inactive') }, { value: 'error', label: t('admin.accounts.status.error') }, { value: 'rate_limited', label: t('admin.accounts.status.rateLimited') }, { value: 'temp_unschedulable', label: t('admin.accounts.status.tempUnschedulable') }])
-const gOpts = computed(() => [{ value: '', label: t('admin.accounts.allGroups') }, ...(props.groups || []).map(g => ({ value: String(g.id), label: g.name }))])
+const gOpts = computed(() => [
+  { value: '', label: t('admin.accounts.allGroups') },
+  { value: 'ungrouped', label: t('admin.accounts.ungroupedGroup') },
+  ...(props.groups || []).map(g => ({ value: String(g.id), label: g.name }))
+])
 </script>

frontend/src/components/admin/announcements/AnnouncementReadStatusDialog.vue

@@ -69,6 +69,7 @@ import { adminAPI } from '@/api/admin'
 import { formatDateTime } from '@/utils/format'
 import type { AnnouncementUserReadStatus } from '@/types'
 import type { Column } from '@/components/common/types'
+import { getPersistedPageSize } from '@/composables/usePersistedPageSize'
 
 import BaseDialog from '@/components/common/BaseDialog.vue'
 import DataTable from '@/components/common/DataTable.vue'
@@ -92,7 +93,7 @@ const search = ref('')
 
 const pagination = reactive({
   page: 1,
-  page_size: 20,
+  page_size: getPersistedPageSize(),
   total: 0,
   pages: 0
 })

frontend/src/components/admin/usage/UsageTable.vue

@@ -25,8 +25,16 @@
           <span class="text-sm text-gray-900 dark:text-white">{{ row.account?.name || '-' }}</span>
         </template>
 
-        <template #cell-model="{ value }">
-          <span class="font-medium text-gray-900 dark:text-white">{{ value }}</span>
+        <template #cell-model="{ row }">
+          <div v-if="row.upstream_model && row.upstream_model !== row.model" class="space-y-0.5 text-xs">
+            <div class="break-all font-medium text-gray-900 dark:text-white">
+              {{ row.model }}
+            </div>
+            <div class="break-all text-gray-500 dark:text-gray-400">
+              <span class="mr-0.5">↳</span>{{ row.upstream_model }}
+            </div>
+          </div>
+          <span v-else class="font-medium text-gray-900 dark:text-white">{{ row.model }}</span>
         </template>
 
         <template #cell-reasoning_effort="{ row }">

frontend/src/components/charts/EndpointDistributionChart.vue

@@ -1,10 +1,10 @@
 <template>
   <div class="card p-4">
-    <div class="mb-4 flex items-start justify-between gap-3">
+    <div class="mb-4 flex items-center justify-between gap-3">
       <h3 class="text-sm font-semibold text-gray-900 dark:text-white">
         {{ title || t('usage.endpointDistribution') }}
       </h3>
-      <div class="flex flex-col items-end gap-2">
+      <div class="flex flex-wrap items-center justify-end gap-2">
         <div
           v-if="showSourceToggle"
           class="inline-flex rounded-lg border border-gray-200 bg-gray-50 p-0.5 dark:border-gray-700 dark:bg-dark-800"

frontend/src/components/charts/ModelDistributionChart.vue

@@ -6,7 +6,42 @@
           ? t('admin.dashboard.modelDistribution')
           : t('admin.dashboard.spendingRankingTitle') }}
       </h3>
-      <div class="flex items-center gap-2">
+      <div class="flex flex-wrap items-center justify-end gap-2">
+        <div
+          v-if="showSourceToggle"
+          class="inline-flex rounded-lg border border-gray-200 bg-gray-50 p-0.5 dark:border-gray-700 dark:bg-dark-800"
+        >
+          <button
+            type="button"
+            class="rounded-md px-2.5 py-1 text-xs font-medium transition-colors"
+            :class="source === 'requested'
+              ? 'bg-white text-gray-900 shadow-sm dark:bg-dark-700 dark:text-white'
+              : 'text-gray-500 hover:text-gray-700 dark:text-gray-400 dark:hover:text-gray-200'"
+            @click="emit('update:source', 'requested')"
+          >
+            {{ t('usage.requestedModel') }}
+          </button>
+          <button
+            type="button"
+            class="rounded-md px-2.5 py-1 text-xs font-medium transition-colors"
+            :class="source === 'upstream'
+              ? 'bg-white text-gray-900 shadow-sm dark:bg-dark-700 dark:text-white'
+              : 'text-gray-500 hover:text-gray-700 dark:text-gray-400 dark:hover:text-gray-200'"
+            @click="emit('update:source', 'upstream')"
+          >
+            {{ t('usage.upstreamModel') }}
+          </button>
+          <button
+            type="button"
+            class="rounded-md px-2.5 py-1 text-xs font-medium transition-colors"
+            :class="source === 'mapping'
+              ? 'bg-white text-gray-900 shadow-sm dark:bg-dark-700 dark:text-white'
+              : 'text-gray-500 hover:text-gray-700 dark:text-gray-400 dark:hover:text-gray-200'"
+            @click="emit('update:source', 'mapping')"
+          >
+            {{ t('usage.mapping') }}
+          </button>
+        </div>
         <div
           v-if="showMetricToggle"
           class="inline-flex rounded-lg border border-gray-200 bg-gray-50 p-0.5 dark:border-gray-700 dark:bg-dark-800"
@@ -215,9 +250,13 @@ ChartJS.register(ArcElement, Tooltip, Legend)
 const { t } = useI18n()
 
 type DistributionMetric = 'tokens' | 'actual_cost'
+type ModelSource = 'requested' | 'upstream' | 'mapping'
 type RankingDisplayItem = UserSpendingRankingItem & { isOther?: boolean }
 const props = withDefaults(defineProps<{
   modelStats: ModelStat[]
+  upstreamModelStats?: ModelStat[]
+  mappingModelStats?: ModelStat[]
+  source?: ModelSource
   enableRankingView?: boolean
   rankingItems?: UserSpendingRankingItem[]
   rankingTotalActualCost?: number
@@ -225,12 +264,16 @@ const props = withDefaults(defineProps<{
   rankingTotalTokens?: number
   loading?: boolean
   metric?: DistributionMetric
+  showSourceToggle?: boolean
   showMetricToggle?: boolean
   rankingLoading?: boolean
   rankingError?: boolean
   startDate?: string
   endDate?: string
 }>(), {
+  upstreamModelStats: () => [],
+  mappingModelStats: () => [],
+  source: 'requested',
   enableRankingView: false,
   rankingItems: () => [],
   rankingTotalActualCost: 0,
@@ -238,6 +281,7 @@ const props = withDefaults(defineProps<{
   rankingTotalTokens: 0,
   loading: false,
   metric: 'tokens',
+  showSourceToggle: false,
   showMetricToggle: false,
   rankingLoading: false,
   rankingError: false
@@ -261,6 +305,7 @@ const toggleBreakdown = async (type: string, id: string) => {
       start_date: props.startDate,
       end_date: props.endDate,
       model: id,
+      model_source: props.source,
     })
     breakdownItems.value = res.users || []
   } catch {
@@ -272,6 +317,7 @@ const toggleBreakdown = async (type: string, id: string) => {
 
 const emit = defineEmits<{
   'update:metric': [value: DistributionMetric]
+  'update:source': [value: ModelSource]
   'ranking-click': [item: UserSpendingRankingItem]
 }>()
 
@@ -294,14 +340,19 @@ const chartColors = [
 ]
 
 const displayModelStats = computed(() => {
-  if (!props.modelStats?.length) return []
+  const sourceStats = props.source === 'upstream'
+    ? props.upstreamModelStats
+    : props.source === 'mapping'
+      ? props.mappingModelStats
+      : props.modelStats
+  if (!sourceStats?.length) return []
 
   const metricKey = props.metric === 'actual_cost' ? 'actual_cost' : 'total_tokens'
-  return [...props.modelStats].sort((a, b) => b[metricKey] - a[metricKey])
+  return [...sourceStats].sort((a, b) => b[metricKey] - a[metricKey])
 })
 
 const chartData = computed(() => {
-  if (!props.modelStats?.length) return null
+  if (!displayModelStats.value.length) return null
 
   return {
     labels: displayModelStats.value.map((m) => m.model),

frontend/src/components/common/DataTable.vue

@@ -147,28 +147,46 @@
           </td>
         </tr>
 
-        <!-- Data rows -->
-        <tr
-          v-else
-          v-for="(row, index) in sortedData"
-          :key="resolveRowKey(row, index)"
-          :data-row-id="resolveRowKey(row, index)"
-          class="hover:bg-gray-50 dark:hover:bg-dark-800"
-        >
-          <td
-            v-for="(column, colIndex) in columns"
-            :key="column.key"
-            :class="[
-              'whitespace-nowrap py-4 text-sm text-gray-900 dark:text-gray-100',
-              getAdaptivePaddingClass(),
-              getStickyColumnClass(column, colIndex)
-            ]"
+        <!-- Data rows (virtual scroll) -->
+        <template v-else>
+          <tr v-if="virtualPaddingTop > 0" aria-hidden="true">
+            <td :colspan="columns.length"
+                :style="{ height: virtualPaddingTop + 'px', padding: 0, border: 'none' }">
+            </td>
+          </tr>
+          <tr
+            v-for="virtualRow in virtualItems"
+            :key="resolveRowKey(sortedData[virtualRow.index], virtualRow.index)"
+            :data-row-id="resolveRowKey(sortedData[virtualRow.index], virtualRow.index)"
+            :data-index="virtualRow.index"
+            :ref="measureElement"
+            class="hover:bg-gray-50 dark:hover:bg-dark-800"
           >
-            <slot :name="`cell-${column.key}`" :row="row" :value="row[column.key]" :expanded="actionsExpanded">
-              {{ column.formatter ? column.formatter(row[column.key], row) : row[column.key] }}
-            </slot>
-          </td>
-        </tr>
+            <td
+              v-for="(column, colIndex) in columns"
+              :key="column.key"
+              :class="[
+                'whitespace-nowrap py-4 text-sm text-gray-900 dark:text-gray-100',
+                getAdaptivePaddingClass(),
+                getStickyColumnClass(column, colIndex)
+              ]"
+            >
+              <slot :name="`cell-${column.key}`"
+                    :row="sortedData[virtualRow.index]"
+                    :value="sortedData[virtualRow.index][column.key]"
+                    :expanded="actionsExpanded">
+                {{ column.formatter
+                   ? column.formatter(sortedData[virtualRow.index][column.key], sortedData[virtualRow.index])
+                   : sortedData[virtualRow.index][column.key] }}
+              </slot>
+            </td>
+          </tr>
+          <tr v-if="virtualPaddingBottom > 0" aria-hidden="true">
+            <td :colspan="columns.length"
+                :style="{ height: virtualPaddingBottom + 'px', padding: 0, border: 'none' }">
+            </td>
+          </tr>
+        </template>
       </tbody>
     </table>
   </div>
@@ -176,6 +194,7 @@
 
 <script setup lang="ts">
 import { computed, ref, onMounted, onUnmounted, watch, nextTick } from 'vue'
+import { useVirtualizer } from '@tanstack/vue-virtual'
 import { useI18n } from 'vue-i18n'
 import type { Column } from './types'
 import Icon from '@/components/icons/Icon.vue'
@@ -299,6 +318,10 @@ interface Props {
    * will emit 'sort' events instead of performing client-side sorting.
    */
   serverSideSort?: boolean
+  /** Estimated row height in px for the virtualizer (default 56) */
+  estimateRowHeight?: number
+  /** Number of rows to render beyond the visible area (default 5) */
+  overscan?: number
 }
 
 const props = withDefaults(defineProps<Props>(), {
@@ -499,6 +522,33 @@ const sortedData = computed(() => {
     .map(item => item.row)
 })
 
+// --- Virtual scrolling ---
+const rowVirtualizer = useVirtualizer(computed(() => ({
+  count: sortedData.value?.length ?? 0,
+  getScrollElement: () => tableWrapperRef.value,
+  estimateSize: () => props.estimateRowHeight ?? 56,
+  overscan: props.overscan ?? 5,
+})))
+
+const virtualItems = computed(() => rowVirtualizer.value.getVirtualItems())
+
+const virtualPaddingTop = computed(() => {
+  const items = virtualItems.value
+  return items.length > 0 ? items[0].start : 0
+})
+
+const virtualPaddingBottom = computed(() => {
+  const items = virtualItems.value
+  if (items.length === 0) return 0
+  return rowVirtualizer.value.getTotalSize() - items[items.length - 1].end
+})
+
+const measureElement = (el: any) => {
+  if (el) {
+    rowVirtualizer.value.measureElement(el as Element)
+  }
+}
+
 const hasActionsColumn = computed(() => {
   return props.columns.some(column => column.key === 'actions')
 })
@@ -595,6 +645,13 @@ watch(
   },
   { flush: 'post' }
 )
+
+defineExpose({
+  virtualizer: rowVirtualizer,
+  sortedData,
+  resolveRowKey,
+  tableWrapperEl: tableWrapperRef,
+})
 </script>
 
 <style scoped>
@@ -603,6 +660,9 @@ watch(
   --select-col-width: 52px; /* 勾选列宽度：px-6 (24px*2) + checkbox (16px) */
   position: relative;
   overflow-x: auto;
+  overflow-y: auto;
+  flex: 1;
+  min-height: 0;
   isolation: isolate;
 }
 

frontend/src/components/common/Pagination.vue

@@ -122,6 +122,7 @@ import { computed, ref } from 'vue'
 import { useI18n } from 'vue-i18n'
 import Icon from '@/components/icons/Icon.vue'
 import Select from './Select.vue'
+import { setPersistedPageSize } from '@/composables/usePersistedPageSize'
 
 const { t } = useI18n()
 
@@ -216,6 +217,7 @@ const goToPage = (newPage: number) => {
 const handlePageSizeChange = (value: string | number | boolean | null) => {
   if (value === null || typeof value === 'boolean') return
   const newPageSize = typeof value === 'string' ? parseInt(value) : value
+  setPersistedPageSize(newPageSize)
   emit('update:pageSize', newPageSize)
 }
 

frontend/src/components/sora/SoraLibraryPage.vue

@@ -126,6 +126,7 @@
 import { ref, computed, onMounted } from 'vue'
 import { useI18n } from 'vue-i18n'
 import soraAPI, { type SoraGeneration } from '@/api/sora'
+import { getPersistedPageSize } from '@/composables/usePersistedPageSize'
 import SoraMediaPreview from './SoraMediaPreview.vue'
 
 const emit = defineEmits<{
@@ -190,7 +191,7 @@ async function loadItems(pageNum: number) {
       status: 'completed',
       storage_type: 's3,local',
       page: pageNum,
-      page_size: 20
+      page_size: getPersistedPageSize()
     })
     const rows = Array.isArray(res.data) ? res.data : []
     if (pageNum === 1) {

frontend/src/composables/usePersistedPageSize.ts

@@ -0,0 +1,27 @@
+const STORAGE_KEY = 'table-page-size'
+const DEFAULT_PAGE_SIZE = 20
+
+/**
+ * 从 localStorage 读取/写入 pageSize
+ * 全局共享一个 key，所有表格统一偏好
+ */
+export function getPersistedPageSize(fallback = DEFAULT_PAGE_SIZE): number {
+  try {
+    const stored = localStorage.getItem(STORAGE_KEY)
+    if (stored) {
+      const parsed = Number(stored)
+      if (Number.isFinite(parsed) && parsed > 0) return parsed
+    }
+  } catch {
+    // localStorage 不可用（隐私模式等）
+  }
+  return fallback
+}
+
+export function setPersistedPageSize(size: number): void {
+  try {
+    localStorage.setItem(STORAGE_KEY, String(size))
+  } catch {
+    // 静默失败
+  }
+}

frontend/src/composables/useSwipeSelect.ts

@@ -1,4 +1,5 @@
 import { ref, onMounted, onUnmounted, type Ref } from 'vue'
+import type { Virtualizer } from '@tanstack/vue-virtual'
 
 /**
  * WeChat-style swipe/drag to select rows in a DataTable,
@@ -25,11 +26,22 @@ export interface SwipeSelectAdapter {
   isSelected: (id: number) => boolean
   select: (id: number) => void
   deselect: (id: number) => void
+  batchUpdate?: (updater: (draft: Set<number>) => void) => void
+}
+
+export interface SwipeSelectVirtualContext {
+  /** Get the virtualizer instance */
+  getVirtualizer: () => Virtualizer<HTMLElement, Element> | null
+  /** Get all sorted data */
+  getSortedData: () => any[]
+  /** Get row ID from data row */
+  getRowId: (row: any, index: number) => number
 }
 
 export function useSwipeSelect(
   containerRef: Ref<HTMLElement | null>,
-  adapter: SwipeSelectAdapter
+  adapter: SwipeSelectAdapter,
+  virtualContext?: SwipeSelectVirtualContext
 ) {
   const isDragging = ref(false)
 
@@ -95,6 +107,32 @@ export function useSwipeSelect(
     return (clientY - rHi.bottom < rLo.top - clientY) ? hi : lo
   }
 
+  /** Virtual mode: find row index from Y coordinate using virtualizer data */
+  function findRowIndexAtYVirtual(clientY: number): number {
+    const virt = virtualContext!.getVirtualizer()
+    if (!virt) return -1
+    const scrollEl = virt.scrollElement
+    if (!scrollEl) return -1
+
+    const scrollRect = scrollEl.getBoundingClientRect()
+    const thead = scrollEl.querySelector('thead')
+    const theadHeight = thead ? thead.getBoundingClientRect().height : 0
+    const contentY = clientY - scrollRect.top - theadHeight + scrollEl.scrollTop
+
+    // Search in rendered virtualItems first
+    const items = virt.getVirtualItems()
+    for (const item of items) {
+      if (contentY >= item.start && contentY < item.end) return item.index
+    }
+
+    // Outside visible range: estimate
+    const totalCount = virtualContext!.getSortedData().length
+    if (totalCount === 0) return -1
+    const est = virt.options.estimateSize(0)
+    const guess = Math.floor(contentY / est)
+    return Math.max(0, Math.min(totalCount - 1, guess))
+  }
+
   // --- Prevent text selection via selectstart (no body style mutation) ---
   function onSelectStart(e: Event) { e.preventDefault() }
 
@@ -140,16 +178,68 @@ export function useSwipeSelect(
     const lo = Math.min(rangeMin, prevMin)
     const hi = Math.max(rangeMax, prevMax)
 
-    for (let i = lo; i <= hi && i < cachedRows.length; i++) {
-      const id = getRowId(cachedRows[i])
-      if (id === null) continue
-      if (i >= rangeMin && i <= rangeMax) {
-        if (dragMode === 'select') adapter.select(id)
-        else adapter.deselect(id)
-      } else {
-        const wasSelected = initialSelectedSnapshot.get(id) ?? false
-        if (wasSelected) adapter.select(id)
-        else adapter.deselect(id)
+    if (adapter.batchUpdate) {
+      adapter.batchUpdate((draft) => {
+        for (let i = lo; i <= hi && i < cachedRows.length; i++) {
+          const id = getRowId(cachedRows[i])
+          if (id === null) continue
+          const shouldBeSelected = (i >= rangeMin && i <= rangeMax)
+            ? (dragMode === 'select')
+            : (initialSelectedSnapshot.get(id) ?? false)
+          if (shouldBeSelected) draft.add(id)
+          else draft.delete(id)
+        }
+      })
+    } else {
+      for (let i = lo; i <= hi && i < cachedRows.length; i++) {
+        const id = getRowId(cachedRows[i])
+        if (id === null) continue
+        if (i >= rangeMin && i <= rangeMax) {
+          if (dragMode === 'select') adapter.select(id)
+          else adapter.deselect(id)
+        } else {
+          const wasSelected = initialSelectedSnapshot.get(id) ?? false
+          if (wasSelected) adapter.select(id)
+          else adapter.deselect(id)
+        }
+      }
+    }
+    lastEndIndex = endIndex
+  }
+
+  /** Virtual mode: apply selection range using data array instead of DOM */
+  function applyRangeVirtual(endIndex: number) {
+    if (startRowIndex < 0 || endIndex < 0) return
+    const rangeMin = Math.min(startRowIndex, endIndex)
+    const rangeMax = Math.max(startRowIndex, endIndex)
+    const prevMin = lastEndIndex >= 0 ? Math.min(startRowIndex, lastEndIndex) : rangeMin
+    const prevMax = lastEndIndex >= 0 ? Math.max(startRowIndex, lastEndIndex) : rangeMax
+    const lo = Math.min(rangeMin, prevMin)
+    const hi = Math.max(rangeMax, prevMax)
+    const data = virtualContext!.getSortedData()
+
+    if (adapter.batchUpdate) {
+      adapter.batchUpdate((draft) => {
+        for (let i = lo; i <= hi && i < data.length; i++) {
+          const id = virtualContext!.getRowId(data[i], i)
+          const shouldBeSelected = (i >= rangeMin && i <= rangeMax)
+            ? (dragMode === 'select')
+            : (initialSelectedSnapshot.get(id) ?? false)
+          if (shouldBeSelected) draft.add(id)
+          else draft.delete(id)
+        }
+      })
+    } else {
+      for (let i = lo; i <= hi && i < data.length; i++) {
+        const id = virtualContext!.getRowId(data[i], i)
+        if (i >= rangeMin && i <= rangeMax) {
+          if (dragMode === 'select') adapter.select(id)
+          else adapter.deselect(id)
+        } else {
+          const wasSelected = initialSelectedSnapshot.get(id) ?? false
+          if (wasSelected) adapter.select(id)
+          else adapter.deselect(id)
+        }
       }
     }
     lastEndIndex = endIndex
@@ -234,8 +324,14 @@ export function useSwipeSelect(
     if (shouldPreferNativeTextSelection(target)) return
     if (shouldPreferNativeSelectionOutsideRows(target)) return
 
-    cachedRows = getDataRows()
-    if (cachedRows.length === 0) return
+    if (virtualContext) {
+      // Virtual mode: check data availability instead of DOM rows
+      const data = virtualContext.getSortedData()
+      if (data.length === 0) return
+    } else {
+      cachedRows = getDataRows()
+      if (cachedRows.length === 0) return
+    }
 
     pendingStartY = e.clientY
     // Prevent text selection as soon as the mouse is down,
@@ -253,13 +349,19 @@ export function useSwipeSelect(
     document.removeEventListener('mousemove', onThresholdMove)
     document.removeEventListener('mouseup', onThresholdUp)
 
-    beginDrag(pendingStartY)
+    if (virtualContext) {
+      beginDragVirtual(pendingStartY)
+    } else {
+      beginDrag(pendingStartY)
+    }
 
     // Process the move that crossed the threshold
     lastMouseY = e.clientY
     updateMarquee(e.clientY)
-    const rowIdx = findRowIndexAtY(e.clientY)
-    if (rowIdx >= 0) applyRange(rowIdx)
+    const findIdx = virtualContext ? findRowIndexAtYVirtual : findRowIndexAtY
+    const apply = virtualContext ? applyRangeVirtual : applyRange
+    const rowIdx = findIdx(e.clientY)
+    if (rowIdx >= 0) apply(rowIdx)
     autoScroll(e)
 
     document.addEventListener('mousemove', onMouseMove)
@@ -306,22 +408,62 @@ export function useSwipeSelect(
     window.getSelection()?.removeAllRanges()
   }
 
+  /** Virtual mode: begin drag using data array */
+  function beginDragVirtual(clientY: number) {
+    startRowIndex = findRowIndexAtYVirtual(clientY)
+    const data = virtualContext!.getSortedData()
+    const startRowId = startRowIndex >= 0 && startRowIndex < data.length
+      ? virtualContext!.getRowId(data[startRowIndex], startRowIndex)
+      : null
+    dragMode = (startRowId !== null && adapter.isSelected(startRowId)) ? 'deselect' : 'select'
+
+    // Build full snapshot from all data rows
+    initialSelectedSnapshot = new Map()
+    for (let i = 0; i < data.length; i++) {
+      const id = virtualContext!.getRowId(data[i], i)
+      initialSelectedSnapshot.set(id, adapter.isSelected(id))
+    }
+
+    isDragging.value = true
+    startY = clientY
+    lastMouseY = clientY
+    lastEndIndex = -1
+
+    // In virtual mode, scroll parent is the virtualizer's scroll element
+    const virt = virtualContext!.getVirtualizer()
+    cachedScrollParent = virt?.scrollElement ?? (containerRef.value ? getScrollParent(containerRef.value) : null)
+
+    createMarquee()
+    updateMarquee(clientY)
+    applyRangeVirtual(startRowIndex)
+    window.getSelection()?.removeAllRanges()
+  }
+
+  let moveRAF = 0
+
   function onMouseMove(e: MouseEvent) {
     if (!isDragging.value) return
     lastMouseY = e.clientY
-    updateMarquee(e.clientY)
-    const rowIdx = findRowIndexAtY(e.clientY)
-    if (rowIdx >= 0 && rowIdx !== lastEndIndex) applyRange(rowIdx)
+    const findIdx = virtualContext ? findRowIndexAtYVirtual : findRowIndexAtY
+    const apply = virtualContext ? applyRangeVirtual : applyRange
+    cancelAnimationFrame(moveRAF)
+    moveRAF = requestAnimationFrame(() => {
+      updateMarquee(lastMouseY)
+      const rowIdx = findIdx(lastMouseY)
+      if (rowIdx >= 0 && rowIdx !== lastEndIndex) apply(rowIdx)
+    })
     autoScroll(e)
   }
 
   function onWheel() {
     if (!isDragging.value) return
+    const findIdx = virtualContext ? findRowIndexAtYVirtual : findRowIndexAtY
+    const apply = virtualContext ? applyRangeVirtual : applyRange
     // After wheel scroll, rows shift in viewport — re-check selection
     requestAnimationFrame(() => {
       if (!isDragging.value) return // guard: drag may have ended before this frame
-      const rowIdx = findRowIndexAtY(lastMouseY)
-      if (rowIdx >= 0) applyRange(rowIdx)
+      const rowIdx = findIdx(lastMouseY)
+      if (rowIdx >= 0) apply(rowIdx)
     })
   }
 
@@ -332,6 +474,7 @@ export function useSwipeSelect(
     cachedRows = []
     initialSelectedSnapshot.clear()
     cachedScrollParent = null
+    cancelAnimationFrame(moveRAF)
     stopAutoScroll()
     removeMarquee()
     document.removeEventListener('selectstart', onSelectStart)
@@ -372,13 +515,15 @@ export function useSwipeSelect(
     }
 
     if (dy !== 0) {
+      const findIdx = virtualContext ? findRowIndexAtYVirtual : findRowIndexAtY
+      const apply = virtualContext ? applyRangeVirtual : applyRange
       const step = () => {
         const prevScrollTop = scrollEl.scrollTop
         scrollEl.scrollTop += dy
         // Only re-check selection if scroll actually moved
         if (scrollEl.scrollTop !== prevScrollTop) {
-          const rowIdx = findRowIndexAtY(lastMouseY)
-          if (rowIdx >= 0 && rowIdx !== lastEndIndex) applyRange(rowIdx)
+          const rowIdx = findIdx(lastMouseY)
+          if (rowIdx >= 0 && rowIdx !== lastEndIndex) apply(rowIdx)
         }
         scrollRAF = requestAnimationFrame(step)
       }

frontend/src/composables/useTableLoader.ts

@@ -1,6 +1,7 @@
 import { ref, reactive, onUnmounted, toRaw } from 'vue'
 import { useDebounceFn } from '@vueuse/core'
 import type { BasePaginationResponse, FetchOptions } from '@/types'
+import { getPersistedPageSize, setPersistedPageSize } from './usePersistedPageSize'
 
 interface PaginationState {
   page: number
@@ -21,14 +22,14 @@ interface TableLoaderOptions<T, P> {
  * 统一处理分页、筛选、搜索防抖和请求取消
  */
 export function useTableLoader<T, P extends Record<string, any>>(options: TableLoaderOptions<T, P>) {
-  const { fetchFn, initialParams, pageSize = 20, debounceMs = 300 } = options
+  const { fetchFn, initialParams, pageSize, debounceMs = 300 } = options
 
   const items = ref<T[]>([])
   const loading = ref(false)
   const params = reactive<P>({ ...(initialParams || {}) } as P)
   const pagination = reactive<PaginationState>({
     page: 1,
-    page_size: pageSize,
+    page_size: pageSize ?? getPersistedPageSize(),
     total: 0,
     pages: 0
   })
@@ -87,6 +88,7 @@ export function useTableLoader<T, P extends Record<string, any>>(options: TableL
   const handlePageSizeChange = (size: number) => {
     pagination.page_size = size
     pagination.page = 1
+    setPersistedPageSize(size)
     load()
   }
 

frontend/src/i18n/locales/en.ts

@@ -218,7 +218,7 @@ export default {
       email: 'Email',
       password: 'Password',
       confirmPassword: 'Confirm Password',
-      passwordPlaceholder: 'Min 6 characters',
+      passwordPlaceholder: 'Min 8 characters',
       confirmPasswordPlaceholder: 'Confirm password',
       passwordMismatch: 'Passwords do not match'
     },
@@ -718,11 +718,14 @@ export default {
     exporting: 'Exporting...',
     preparingExport: 'Preparing export...',
     model: 'Model',
+    requestedModel: 'Requested',
+    upstreamModel: 'Upstream',
     reasoningEffort: 'Reasoning Effort',
     endpoint: 'Endpoint',
     endpointDistribution: 'Endpoint Distribution',
     inbound: 'Inbound',
     upstream: 'Upstream',
+    mapping: 'Mapping',
     path: 'Path',
     inboundEndpoint: 'Inbound Endpoint',
     upstreamEndpoint: 'Upstream Endpoint',
@@ -1880,6 +1883,7 @@ export default {
       allTypes: 'All Types',
       allStatus: 'All Status',
       allGroups: 'All Groups',
+      ungroupedGroup: 'Ungrouped',
       oauthType: 'OAuth',
       setupToken: 'Setup Token',
       apiKey: 'API Key',
@@ -2757,7 +2761,9 @@ export default {
         gemini3Pro: 'G3P',
         gemini3Flash: 'G3F',
         gemini3Image: 'G31FI',
-        claude: 'Claude'
+        claude: 'Claude',
+        passiveSampled: 'Passive',
+        activeQuery: 'Query'
       },
       tier: {
         free: 'Free',
@@ -4359,6 +4365,16 @@ export default {
           testFailed: 'Google Drive storage test failed'
         }
       },
+      overloadCooldown: {
+        title: '529 Overload Cooldown',
+        description: 'Configure account scheduling pause strategy when upstream returns 529 (overloaded)',
+        enabled: 'Enable Overload Cooldown',
+        enabledHint: 'Pause account scheduling on 529 errors, auto-recover after cooldown',
+        cooldownMinutes: 'Cooldown Duration (minutes)',
+        cooldownMinutesHint: 'Duration to pause account scheduling (1-120 minutes)',
+        saved: 'Overload cooldown settings saved',
+        saveFailed: 'Failed to save overload cooldown settings'
+      },
       streamTimeout: {
         title: 'Stream Timeout Handling',
         description: 'Configure account handling strategy when upstream response times out',

frontend/src/i18n/locales/zh.ts

@@ -218,7 +218,7 @@ export default {
       email: '邮箱',
       password: '密码',
       confirmPassword: '确认密码',
-      passwordPlaceholder: '至少 6 个字符',
+      passwordPlaceholder: '至少 8 个字符',
       confirmPasswordPlaceholder: '确认密码',
       passwordMismatch: '密码不匹配'
     },
@@ -723,11 +723,14 @@ export default {
     exporting: '导出中...',
     preparingExport: '正在准备导出...',
     model: '模型',
+    requestedModel: '请求',
+    upstreamModel: '上游',
     reasoningEffort: '推理强度',
     endpoint: '端点',
     endpointDistribution: '端点分布',
     inbound: '入站',
     upstream: '上游',
+    mapping: '映射',
     path: '路径',
     inboundEndpoint: '入站端点',
     upstreamEndpoint: '上游端点',
@@ -1962,6 +1965,7 @@ export default {
       allTypes: '全部类型',
       allStatus: '全部状态',
       allGroups: '全部分组',
+      ungroupedGroup: '未分配分组',
       oauthType: 'OAuth',
       // Schedulable toggle
       schedulable: '参与调度',
@@ -2160,7 +2164,9 @@ export default {
         gemini3Pro: 'G3P',
         gemini3Flash: 'G3F',
         gemini3Image: 'G31FI',
-        claude: 'Claude'
+        claude: 'Claude',
+        passiveSampled: '被动采样',
+        activeQuery: '查询'
       },
       tier: {
         free: 'Free',
@@ -4524,6 +4530,16 @@ export default {
           testFailed: 'Google Drive 存储测试失败'
         }
       },
+      overloadCooldown: {
+        title: '529 过载冷却',
+        description: '配置上游返回 529（过载）时的账号调度暂停策略',
+        enabled: '启用过载冷却',
+        enabledHint: '收到 529 错误时暂停该账号的调度，冷却后自动恢复',
+        cooldownMinutes: '冷却时长（分钟）',
+        cooldownMinutesHint: '账号暂停调度的持续时间（1-120 分钟）',
+        saved: '过载冷却设置保存成功',
+        saveFailed: '保存过载冷却设置失败'
+      },
       streamTimeout: {
         title: '流超时处理',
         description: '配置上游响应超时时的账户处理策略，避免问题账户持续被选中',

frontend/src/types/index.ts

@@ -781,6 +781,7 @@ export interface AntigravityModelQuota {
 }
 
 export interface AccountUsageInfo {
+  source?: 'passive' | 'active'
   updated_at: string | null
   five_hour: UsageProgress | null
   seven_day: UsageProgress | null
@@ -977,6 +978,7 @@ export interface UsageLog {
   account_id: number | null
   request_id: string
   model: string
+  upstream_model?: string | null
   service_tier?: string | null
   reasoning_effort?: string | null
   inbound_endpoint?: string | null

frontend/src/views/admin/AccountsView.vue

@@ -758,6 +758,7 @@ const refreshAccountsIncrementally = async () => {
         platform?: string
         type?: string
         status?: string
+        group?: string
         search?: string
 
       },

frontend/src/views/admin/AnnouncementsView.vue

@@ -239,6 +239,7 @@
 import { computed, onMounted, reactive, ref } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { useAppStore } from '@/stores/app'
+import { getPersistedPageSize } from '@/composables/usePersistedPageSize'
 import { adminAPI } from '@/api/admin'
 import { formatDateTime, formatDateTimeLocalInput, parseDateTimeLocalInput } from '@/utils/format'
 import type { AdminGroup, Announcement, AnnouncementTargeting } from '@/types'
@@ -270,7 +271,7 @@ const searchQuery = ref('')
 
 const pagination = reactive({
   page: 1,
-  page_size: 20,
+  page_size: getPersistedPageSize(),
   total: 0,
   pages: 0
 })

frontend/src/views/admin/GroupsView.vue

@@ -1855,6 +1855,7 @@ import GroupCapacityBadge from '@/components/common/GroupCapacityBadge.vue'
 import { VueDraggable } from 'vue-draggable-plus'
 import { createStableObjectKeyResolver } from '@/utils/stableObjectKey'
 import { useKeyedDebouncedSearch } from '@/composables/useKeyedDebouncedSearch'
+import { getPersistedPageSize } from '@/composables/usePersistedPageSize'
 
 const { t } = useI18n()
 const appStore = useAppStore()
@@ -2016,7 +2017,7 @@ const filters = reactive({
 })
 const pagination = reactive({
   page: 1,
-  page_size: 20,
+  page_size: getPersistedPageSize(),
   total: 0,
   pages: 0
 })

frontend/src/views/admin/PromoCodesView.vue

@@ -383,6 +383,7 @@ import { ref, reactive, computed, onMounted, onUnmounted } from 'vue'
 import { useI18n } from 'vue-i18n'
 import { useAppStore } from '@/stores/app'
 import { useClipboard } from '@/composables/useClipboard'
+import { getPersistedPageSize } from '@/composables/usePersistedPageSize'
 import { adminAPI } from '@/api/admin'
 import { formatDateTime } from '@/utils/format'
 import type { PromoCode, PromoCodeUsage } from '@/types'
@@ -414,7 +415,7 @@ const filters = reactive({
 
 const pagination = reactive({
   page: 1,
-  page_size: 20,
+  page_size: getPersistedPageSize(),
   total: 0
 })