feat: add claude max usage simulation with group switch

backend/internal/handler/admin/group_handler.go

@@ -46,9 +46,10 @@ type CreateGroupRequest struct {
 	FallbackGroupID                 *int64   `json:"fallback_group_id"`
 	FallbackGroupIDOnInvalidRequest *int64   `json:"fallback_group_id_on_invalid_request"`
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64 `json:"model_routing"`
-	ModelRoutingEnabled bool               `json:"model_routing_enabled"`
-	MCPXMLInject        *bool              `json:"mcp_xml_inject"`
+	ModelRouting             map[string][]int64 `json:"model_routing"`
+	ModelRoutingEnabled      bool               `json:"model_routing_enabled"`
+	MCPXMLInject             *bool              `json:"mcp_xml_inject"`
+	SimulateClaudeMaxEnabled *bool              `json:"simulate_claude_max_enabled"`
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string `json:"supported_model_scopes"`
 	// 从指定分组复制账号（创建后自动绑定）
@@ -79,9 +80,10 @@ type UpdateGroupRequest struct {
 	FallbackGroupID                 *int64   `json:"fallback_group_id"`
 	FallbackGroupIDOnInvalidRequest *int64   `json:"fallback_group_id_on_invalid_request"`
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64 `json:"model_routing"`
-	ModelRoutingEnabled *bool              `json:"model_routing_enabled"`
-	MCPXMLInject        *bool              `json:"mcp_xml_inject"`
+	ModelRouting             map[string][]int64 `json:"model_routing"`
+	ModelRoutingEnabled      *bool              `json:"model_routing_enabled"`
+	MCPXMLInject             *bool              `json:"mcp_xml_inject"`
+	SimulateClaudeMaxEnabled *bool              `json:"simulate_claude_max_enabled"`
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes *[]string `json:"supported_model_scopes"`
 	// 从指定分组复制账号（同步操作：先清空当前分组的账号绑定，再绑定源分组的账号）
@@ -197,6 +199,7 @@ func (h *GroupHandler) Create(c *gin.Context) {
 		ModelRouting:                    req.ModelRouting,
 		ModelRoutingEnabled:             req.ModelRoutingEnabled,
 		MCPXMLInject:                    req.MCPXMLInject,
+		SimulateClaudeMaxEnabled:        req.SimulateClaudeMaxEnabled,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
 	})
@@ -247,6 +250,7 @@ func (h *GroupHandler) Update(c *gin.Context) {
 		ModelRouting:                    req.ModelRouting,
 		ModelRoutingEnabled:             req.ModelRoutingEnabled,
 		MCPXMLInject:                    req.MCPXMLInject,
+		SimulateClaudeMaxEnabled:        req.SimulateClaudeMaxEnabled,
 		SupportedModelScopes:            req.SupportedModelScopes,
 		CopyAccountsFromGroupIDs:        req.CopyAccountsFromGroupIDs,
 	})

backend/internal/handler/dto/mappers.go

@@ -111,13 +111,14 @@ func GroupFromServiceAdmin(g *service.Group) *AdminGroup {
 		return nil
 	}
 	out := &AdminGroup{
-		Group:                groupFromServiceBase(g),
-		ModelRouting:         g.ModelRouting,
-		ModelRoutingEnabled:  g.ModelRoutingEnabled,
-		MCPXMLInject:         g.MCPXMLInject,
-		SupportedModelScopes: g.SupportedModelScopes,
-		AccountCount:         g.AccountCount,
-		SortOrder:            g.SortOrder,
+		Group:                    groupFromServiceBase(g),
+		ModelRouting:             g.ModelRouting,
+		ModelRoutingEnabled:      g.ModelRoutingEnabled,
+		MCPXMLInject:             g.MCPXMLInject,
+		SimulateClaudeMaxEnabled: g.SimulateClaudeMaxEnabled,
+		SupportedModelScopes:     g.SupportedModelScopes,
+		AccountCount:             g.AccountCount,
+		SortOrder:                g.SortOrder,
 	}
 	if len(g.AccountGroups) > 0 {
 		out.AccountGroups = make([]AccountGroup, 0, len(g.AccountGroups))

backend/internal/handler/dto/types.go

@@ -95,6 +95,8 @@ type AdminGroup struct {
 
 	// MCP XML 协议注入（仅 antigravity 平台使用）
 	MCPXMLInject bool `json:"mcp_xml_inject"`
+	// Claude usage 模拟开关（仅管理员可见）
+	SimulateClaudeMaxEnabled bool `json:"simulate_claude_max_enabled"`
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string       `json:"supported_model_scopes"`

backend/internal/handler/gateway_handler.go

@@ -405,6 +405,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			h.submitUsageRecordTask(func(ctx context.Context) {
 				if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
 					Result:            result,
+					ParsedRequest:     parsedReq,
 					APIKey:            apiKey,
 					User:              apiKey.User,
 					Account:           account,
@@ -631,6 +632,7 @@ func (h *GatewayHandler) Messages(c *gin.Context) {
 			h.submitUsageRecordTask(func(ctx context.Context) {
 				if err := h.gatewayService.RecordUsage(ctx, &service.RecordUsageInput{
 					Result:            result,
+					ParsedRequest:     parsedReq,
 					APIKey:            currentAPIKey,
 					User:              currentAPIKey.User,
 					Account:           account,

backend/internal/repository/api_key_repo.go

@@ -152,6 +152,7 @@ func (r *apiKeyRepository) GetByKeyForAuth(ctx context.Context, key string) (*se
 				group.FieldModelRoutingEnabled,
 				group.FieldModelRouting,
 				group.FieldMcpXMLInject,
+				group.FieldSimulateClaudeMaxEnabled,
 				group.FieldSupportedModelScopes,
 			)
 		}).
@@ -493,6 +494,7 @@ func groupEntityToService(g *dbent.Group) *service.Group {
 		ModelRouting:                    g.ModelRouting,
 		ModelRoutingEnabled:             g.ModelRoutingEnabled,
 		MCPXMLInject:                    g.McpXMLInject,
+		SimulateClaudeMaxEnabled:        g.SimulateClaudeMaxEnabled,
 		SupportedModelScopes:            g.SupportedModelScopes,
 		SortOrder:                       g.SortOrder,
 		CreatedAt:                       g.CreatedAt,

backend/internal/repository/group_repo.go

@@ -56,7 +56,8 @@ func (r *groupRepository) Create(ctx context.Context, groupIn *service.Group) er
 		SetNillableFallbackGroupID(groupIn.FallbackGroupID).
 		SetNillableFallbackGroupIDOnInvalidRequest(groupIn.FallbackGroupIDOnInvalidRequest).
 		SetModelRoutingEnabled(groupIn.ModelRoutingEnabled).
-		SetMcpXMLInject(groupIn.MCPXMLInject)
+		SetMcpXMLInject(groupIn.MCPXMLInject).
+		SetSimulateClaudeMaxEnabled(groupIn.SimulateClaudeMaxEnabled)
 
 	// 设置模型路由配置
 	if groupIn.ModelRouting != nil {
@@ -121,7 +122,8 @@ func (r *groupRepository) Update(ctx context.Context, groupIn *service.Group) er
 		SetDefaultValidityDays(groupIn.DefaultValidityDays).
 		SetClaudeCodeOnly(groupIn.ClaudeCodeOnly).
 		SetModelRoutingEnabled(groupIn.ModelRoutingEnabled).
-		SetMcpXMLInject(groupIn.MCPXMLInject)
+		SetMcpXMLInject(groupIn.MCPXMLInject).
+		SetSimulateClaudeMaxEnabled(groupIn.SimulateClaudeMaxEnabled)
 
 	// 处理 FallbackGroupID：nil 时清除，否则设置
 	if groupIn.FallbackGroupID != nil {

backend/internal/service/admin_service.go

@@ -130,9 +130,10 @@ type CreateGroupInput struct {
 	// 无效请求兜底分组 ID（仅 anthropic 平台使用）
 	FallbackGroupIDOnInvalidRequest *int64
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64
-	ModelRoutingEnabled bool // 是否启用模型路由
-	MCPXMLInject        *bool
+	ModelRouting             map[string][]int64
+	ModelRoutingEnabled      bool // 是否启用模型路由
+	MCPXMLInject             *bool
+	SimulateClaudeMaxEnabled *bool
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string
 	// 从指定分组复制账号（创建分组后在同一事务内绑定）
@@ -164,9 +165,10 @@ type UpdateGroupInput struct {
 	// 无效请求兜底分组 ID（仅 anthropic 平台使用）
 	FallbackGroupIDOnInvalidRequest *int64
 	// 模型路由配置（仅 anthropic 平台使用）
-	ModelRouting        map[string][]int64
-	ModelRoutingEnabled *bool // 是否启用模型路由
-	MCPXMLInject        *bool
+	ModelRouting             map[string][]int64
+	ModelRoutingEnabled      *bool // 是否启用模型路由
+	MCPXMLInject             *bool
+	SimulateClaudeMaxEnabled *bool
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes *[]string
 	// 从指定分组复制账号（同步操作：先清空当前分组的账号绑定，再绑定源分组的账号）
@@ -763,6 +765,13 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 	if input.MCPXMLInject != nil {
 		mcpXMLInject = *input.MCPXMLInject
 	}
+	simulateClaudeMaxEnabled := false
+	if input.SimulateClaudeMaxEnabled != nil {
+		if platform != PlatformAnthropic && *input.SimulateClaudeMaxEnabled {
+			return nil, fmt.Errorf("simulate_claude_max_enabled only supported for anthropic groups")
+		}
+		simulateClaudeMaxEnabled = *input.SimulateClaudeMaxEnabled
+	}
 
 	// 如果指定了复制账号的源分组，先获取账号 ID 列表
 	var accountIDsToCopy []int64
@@ -819,6 +828,7 @@ func (s *adminServiceImpl) CreateGroup(ctx context.Context, input *CreateGroupIn
 		FallbackGroupIDOnInvalidRequest: fallbackOnInvalidRequest,
 		ModelRouting:                    input.ModelRouting,
 		MCPXMLInject:                    mcpXMLInject,
+		SimulateClaudeMaxEnabled:        simulateClaudeMaxEnabled,
 		SupportedModelScopes:            input.SupportedModelScopes,
 	}
 	if err := s.groupRepo.Create(ctx, group); err != nil {
@@ -1024,6 +1034,15 @@ func (s *adminServiceImpl) UpdateGroup(ctx context.Context, id int64, input *Upd
 	if input.MCPXMLInject != nil {
 		group.MCPXMLInject = *input.MCPXMLInject
 	}
+	if input.SimulateClaudeMaxEnabled != nil {
+		if group.Platform != PlatformAnthropic && *input.SimulateClaudeMaxEnabled {
+			return nil, fmt.Errorf("simulate_claude_max_enabled only supported for anthropic groups")
+		}
+		group.SimulateClaudeMaxEnabled = *input.SimulateClaudeMaxEnabled
+	}
+	if group.Platform != PlatformAnthropic {
+		group.SimulateClaudeMaxEnabled = false
+	}
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	if input.SupportedModelScopes != nil {

backend/internal/service/admin_service_group_test.go

@@ -785,3 +785,57 @@ func TestAdminService_UpdateGroup_InvalidRequestFallbackAllowsAntigravity(t *tes
 	require.NotNil(t, repo.updated)
 	require.Equal(t, fallbackID, *repo.updated.FallbackGroupIDOnInvalidRequest)
 }
+
+func TestAdminService_CreateGroup_SimulateClaudeMaxRequiresAnthropic(t *testing.T) {
+	repo := &groupRepoStubForAdmin{}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	enabled := true
+	_, err := svc.CreateGroup(context.Background(), &CreateGroupInput{
+		Name:                     "openai-group",
+		Platform:                 PlatformOpenAI,
+		SimulateClaudeMaxEnabled: &enabled,
+	})
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "simulate_claude_max_enabled only supported for anthropic groups")
+	require.Nil(t, repo.created)
+}
+
+func TestAdminService_UpdateGroup_SimulateClaudeMaxRequiresAnthropic(t *testing.T) {
+	existingGroup := &Group{
+		ID:       1,
+		Name:     "openai-group",
+		Platform: PlatformOpenAI,
+		Status:   StatusActive,
+	}
+	repo := &groupRepoStubForAdmin{getByID: existingGroup}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	enabled := true
+	_, err := svc.UpdateGroup(context.Background(), 1, &UpdateGroupInput{
+		SimulateClaudeMaxEnabled: &enabled,
+	})
+	require.Error(t, err)
+	require.Contains(t, err.Error(), "simulate_claude_max_enabled only supported for anthropic groups")
+	require.Nil(t, repo.updated)
+}
+
+func TestAdminService_UpdateGroup_ClearsSimulateClaudeMaxWhenPlatformChanges(t *testing.T) {
+	existingGroup := &Group{
+		ID:                       1,
+		Name:                     "anthropic-group",
+		Platform:                 PlatformAnthropic,
+		Status:                   StatusActive,
+		SimulateClaudeMaxEnabled: true,
+	}
+	repo := &groupRepoStubForAdmin{getByID: existingGroup}
+	svc := &adminServiceImpl{groupRepo: repo}
+
+	group, err := svc.UpdateGroup(context.Background(), 1, &UpdateGroupInput{
+		Platform: PlatformOpenAI,
+	})
+	require.NoError(t, err)
+	require.NotNil(t, group)
+	require.NotNil(t, repo.updated)
+	require.False(t, repo.updated.SimulateClaudeMaxEnabled)
+}

backend/internal/service/api_key_auth_cache.go

@@ -54,9 +54,10 @@ type APIKeyAuthGroupSnapshot struct {
 
 	// Model routing is used by gateway account selection, so it must be part of auth cache snapshot.
 	// Only anthropic groups use these fields; others may leave them empty.
-	ModelRouting        map[string][]int64 `json:"model_routing,omitempty"`
-	ModelRoutingEnabled bool               `json:"model_routing_enabled"`
-	MCPXMLInject        bool               `json:"mcp_xml_inject"`
+	ModelRouting             map[string][]int64 `json:"model_routing,omitempty"`
+	ModelRoutingEnabled      bool               `json:"model_routing_enabled"`
+	MCPXMLInject             bool               `json:"mcp_xml_inject"`
+	SimulateClaudeMaxEnabled bool               `json:"simulate_claude_max_enabled"`
 
 	// 支持的模型系列（仅 antigravity 平台使用）
 	SupportedModelScopes []string `json:"supported_model_scopes,omitempty"`

backend/internal/service/api_key_auth_cache_impl.go

@@ -241,6 +241,7 @@ func (s *APIKeyService) snapshotFromAPIKey(apiKey *APIKey) *APIKeyAuthSnapshot {
 			ModelRouting:                    apiKey.Group.ModelRouting,
 			ModelRoutingEnabled:             apiKey.Group.ModelRoutingEnabled,
 			MCPXMLInject:                    apiKey.Group.MCPXMLInject,
+			SimulateClaudeMaxEnabled:        apiKey.Group.SimulateClaudeMaxEnabled,
 			SupportedModelScopes:            apiKey.Group.SupportedModelScopes,
 		}
 	}
@@ -295,6 +296,7 @@ func (s *APIKeyService) snapshotToAPIKey(key string, snapshot *APIKeyAuthSnapsho
 			ModelRouting:                    snapshot.Group.ModelRouting,
 			ModelRoutingEnabled:             snapshot.Group.ModelRoutingEnabled,
 			MCPXMLInject:                    snapshot.Group.MCPXMLInject,
+			SimulateClaudeMaxEnabled:        snapshot.Group.SimulateClaudeMaxEnabled,
 			SupportedModelScopes:            snapshot.Group.SupportedModelScopes,
 		}
 	}

backend/internal/service/claude_max_simulation_test.go

@@ -0,0 +1,92 @@
+package service
+
+import "testing"
+
+func TestProjectUsageToClaudeMax1H_Conservation(t *testing.T) {
+	usage := &ClaudeUsage{
+		InputTokens:              1200,
+		CacheCreationInputTokens: 0,
+		CacheCreation5mTokens:    0,
+		CacheCreation1hTokens:    0,
+	}
+	parsed := &ParsedRequest{
+		Model: "claude-sonnet-4-5",
+		Messages: []any{
+			map[string]any{
+				"role":    "user",
+				"content": "请帮我总结这段代码并给出优化建议",
+			},
+		},
+	}
+
+	changed := projectUsageToClaudeMax1H(usage, parsed)
+	if !changed {
+		t.Fatalf("expected usage to be projected")
+	}
+
+	total := usage.InputTokens + usage.CacheCreation5mTokens + usage.CacheCreation1hTokens
+	if total != 1200 {
+		t.Fatalf("total tokens changed: got=%d want=%d", total, 1200)
+	}
+	if usage.CacheCreation5mTokens != 0 {
+		t.Fatalf("cache_creation_5m should be 0, got=%d", usage.CacheCreation5mTokens)
+	}
+	if usage.InputTokens <= 0 || usage.InputTokens >= 1200 {
+		t.Fatalf("simulated input out of range, got=%d", usage.InputTokens)
+	}
+	if usage.CacheCreation1hTokens <= 0 {
+		t.Fatalf("cache_creation_1h should be > 0, got=%d", usage.CacheCreation1hTokens)
+	}
+	if usage.CacheCreationInputTokens != usage.CacheCreation1hTokens {
+		t.Fatalf("cache_creation_input_tokens mismatch: got=%d want=%d", usage.CacheCreationInputTokens, usage.CacheCreation1hTokens)
+	}
+}
+
+func TestComputeClaudeMaxSimulatedInputTokens_Deterministic(t *testing.T) {
+	parsed := &ParsedRequest{
+		Model: "claude-opus-4-5",
+		Messages: []any{
+			map[string]any{
+				"role": "user",
+				"content": []any{
+					map[string]any{"type": "text", "text": "请整理以下日志并定位错误根因"},
+					map[string]any{"type": "tool_use", "name": "grep_logs"},
+				},
+			},
+		},
+	}
+
+	got1 := computeClaudeMaxSimulatedInputTokens(4096, parsed)
+	got2 := computeClaudeMaxSimulatedInputTokens(4096, parsed)
+	if got1 != got2 {
+		t.Fatalf("non-deterministic input tokens: %d != %d", got1, got2)
+	}
+}
+
+func TestShouldSimulateClaudeMaxUsage(t *testing.T) {
+	group := &Group{
+		Platform:                 PlatformAnthropic,
+		SimulateClaudeMaxEnabled: true,
+	}
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			Model: "claude-sonnet-4-5",
+			Usage: ClaudeUsage{
+				InputTokens:              3000,
+				CacheCreationInputTokens: 0,
+				CacheCreation5mTokens:    0,
+				CacheCreation1hTokens:    0,
+			},
+		},
+		APIKey: &APIKey{Group: group},
+	}
+
+	if !shouldSimulateClaudeMaxUsage(input) {
+		t.Fatalf("expected simulate=true for claude group without cache creation")
+	}
+
+	input.Result.Usage.CacheCreationInputTokens = 100
+	if shouldSimulateClaudeMaxUsage(input) {
+		t.Fatalf("expected simulate=false when cache creation already exists")
+	}
+}

backend/internal/service/gateway_record_usage_claude_max_test.go

@@ -0,0 +1,140 @@
+package service
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/stretchr/testify/require"
+)
+
+type usageLogRepoRecordUsageStub struct {
+	UsageLogRepository
+
+	last     *UsageLog
+	inserted bool
+	err      error
+}
+
+func (s *usageLogRepoRecordUsageStub) Create(_ context.Context, log *UsageLog) (bool, error) {
+	copied := *log
+	s.last = &copied
+	return s.inserted, s.err
+}
+
+func newGatewayServiceForRecordUsageTest(repo UsageLogRepository) *GatewayService {
+	return &GatewayService{
+		usageLogRepo:    repo,
+		billingService:  NewBillingService(&config.Config{}, nil),
+		cfg:             &config.Config{RunMode: config.RunModeSimple},
+		deferredService: &DeferredService{},
+	}
+}
+
+func TestRecordUsage_SimulateClaudeMaxEnabled_ProjectsAndSkipsTTLOverride(t *testing.T) {
+	repo := &usageLogRepoRecordUsageStub{inserted: true}
+	svc := newGatewayServiceForRecordUsageTest(repo)
+
+	groupID := int64(11)
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			RequestID: "req-sim-1",
+			Model:     "claude-sonnet-4",
+			Duration:  time.Second,
+			Usage: ClaudeUsage{
+				InputTokens: 160,
+			},
+		},
+		ParsedRequest: &ParsedRequest{
+			Model: "claude-sonnet-4",
+			Messages: []any{
+				map[string]any{
+					"role":    "user",
+					"content": "please summarize the logs and provide root cause analysis",
+				},
+			},
+		},
+		APIKey: &APIKey{
+			ID:      1,
+			GroupID: &groupID,
+			Group: &Group{
+				ID:                       groupID,
+				Platform:                 PlatformAnthropic,
+				RateMultiplier:           1,
+				SimulateClaudeMaxEnabled: true,
+			},
+		},
+		User: &User{ID: 2},
+		Account: &Account{
+			ID:       3,
+			Platform: PlatformAnthropic,
+			Type:     AccountTypeOAuth,
+			Extra: map[string]any{
+				"cache_ttl_override_enabled": true,
+				"cache_ttl_override_target":  "5m",
+			},
+		},
+	}
+
+	err := svc.RecordUsage(context.Background(), input)
+	require.NoError(t, err)
+	require.NotNil(t, repo.last)
+
+	log := repo.last
+	total := log.InputTokens + log.CacheCreation5mTokens + log.CacheCreation1hTokens
+	require.Equal(t, 160, total, "token 总量应保持不变")
+	require.Greater(t, log.CacheCreation1hTokens, 0, "应映射为 1h cache creation")
+	require.Equal(t, 0, log.CacheCreation5mTokens, "模拟成功后不应再被 TTL override 改写为 5m")
+	require.Equal(t, log.CacheCreation1hTokens, log.CacheCreationTokens, "聚合 cache_creation_tokens 应与 1h 一致")
+	require.False(t, log.CacheTTLOverridden, "模拟成功时应跳过 TTL override 标记")
+}
+
+func TestRecordUsage_SimulateClaudeMaxDisabled_AppliesTTLOverride(t *testing.T) {
+	repo := &usageLogRepoRecordUsageStub{inserted: true}
+	svc := newGatewayServiceForRecordUsageTest(repo)
+
+	groupID := int64(12)
+	input := &RecordUsageInput{
+		Result: &ForwardResult{
+			RequestID: "req-sim-2",
+			Model:     "claude-sonnet-4",
+			Duration:  time.Second,
+			Usage: ClaudeUsage{
+				InputTokens:              40,
+				CacheCreationInputTokens: 120,
+				CacheCreation1hTokens:    120,
+			},
+		},
+		APIKey: &APIKey{
+			ID:      2,
+			GroupID: &groupID,
+			Group: &Group{
+				ID:                       groupID,
+				Platform:                 PlatformAnthropic,
+				RateMultiplier:           1,
+				SimulateClaudeMaxEnabled: false,
+			},
+		},
+		User: &User{ID: 3},
+		Account: &Account{
+			ID:       4,
+			Platform: PlatformAnthropic,
+			Type:     AccountTypeOAuth,
+			Extra: map[string]any{
+				"cache_ttl_override_enabled": true,
+				"cache_ttl_override_target":  "5m",
+			},
+		},
+	}
+
+	err := svc.RecordUsage(context.Background(), input)
+	require.NoError(t, err)
+	require.NotNil(t, repo.last)
+
+	log := repo.last
+	require.Equal(t, 120, log.CacheCreationTokens)
+	require.Equal(t, 120, log.CacheCreation5mTokens, "关闭模拟时应执行 TTL override 到 5m")
+	require.Equal(t, 0, log.CacheCreation1hTokens)
+	require.True(t, log.CacheTTLOverridden, "TTL override 生效时应打标")
+}

backend/internal/service/gateway_service.go

@@ -56,6 +56,15 @@ const (
 	claudeMimicDebugInfoKey = "claude_mimic_debug_info"
 )
 
+const (
+	claudeMaxSimInputMinTokens     = 8
+	claudeMaxSimInputMaxTokens     = 96
+	claudeMaxSimBaseOverheadTokens = 8
+	claudeMaxSimPerBlockOverhead   = 2
+	claudeMaxSimSummaryMaxRunes    = 160
+	claudeMaxSimContextDivisor     = 16
+)
+
 // ForceCacheBillingContextKey 强制缓存计费上下文键
 // 用于粘性会话切换时，将 input_tokens 转为 cache_read_input_tokens 计费
 type forceCacheBillingKeyType struct{}
@@ -5566,9 +5575,228 @@ func (s *GatewayService) getUserGroupRateMultiplier(ctx context.Context, userID,
 	return multiplier
 }
 
+func isClaudeFamilyModel(model string) bool {
+	normalized := strings.ToLower(strings.TrimSpace(claude.NormalizeModelID(model)))
+	if normalized == "" {
+		return false
+	}
+	return strings.Contains(normalized, "claude-")
+}
+
+func shouldSimulateClaudeMaxUsage(input *RecordUsageInput) bool {
+	if input == nil || input.Result == nil || input.APIKey == nil || input.APIKey.Group == nil {
+		return false
+	}
+	group := input.APIKey.Group
+	if !group.SimulateClaudeMaxEnabled || group.Platform != PlatformAnthropic {
+		return false
+	}
+
+	model := input.Result.Model
+	if model == "" && input.ParsedRequest != nil {
+		model = input.ParsedRequest.Model
+	}
+	if !isClaudeFamilyModel(model) {
+		return false
+	}
+
+	usage := input.Result.Usage
+	if usage.InputTokens <= 0 {
+		return false
+	}
+	if usage.CacheCreationInputTokens > 0 || usage.CacheCreation5mTokens > 0 || usage.CacheCreation1hTokens > 0 {
+		return false
+	}
+	return true
+}
+
+func applyClaudeMaxUsageSimulation(result *ForwardResult, parsed *ParsedRequest) bool {
+	if result == nil {
+		return false
+	}
+	return projectUsageToClaudeMax1H(&result.Usage, parsed)
+}
+
+func projectUsageToClaudeMax1H(usage *ClaudeUsage, parsed *ParsedRequest) bool {
+	if usage == nil {
+		return false
+	}
+	totalWindowTokens := usage.InputTokens + usage.CacheCreation5mTokens + usage.CacheCreation1hTokens
+	if totalWindowTokens <= 1 {
+		return false
+	}
+
+	simulatedInputTokens := computeClaudeMaxSimulatedInputTokens(totalWindowTokens, parsed)
+	if simulatedInputTokens <= 0 {
+		simulatedInputTokens = 1
+	}
+	if simulatedInputTokens >= totalWindowTokens {
+		simulatedInputTokens = totalWindowTokens - 1
+	}
+
+	cacheCreation1hTokens := totalWindowTokens - simulatedInputTokens
+	if usage.InputTokens == simulatedInputTokens &&
+		usage.CacheCreation5mTokens == 0 &&
+		usage.CacheCreation1hTokens == cacheCreation1hTokens &&
+		usage.CacheCreationInputTokens == cacheCreation1hTokens {
+		return false
+	}
+
+	usage.InputTokens = simulatedInputTokens
+	usage.CacheCreation5mTokens = 0
+	usage.CacheCreation1hTokens = cacheCreation1hTokens
+	usage.CacheCreationInputTokens = cacheCreation1hTokens
+	return true
+}
+
+func computeClaudeMaxSimulatedInputTokens(totalWindowTokens int, parsed *ParsedRequest) int {
+	if totalWindowTokens <= 1 {
+		return totalWindowTokens
+	}
+
+	summary, blockCount := extractTailUserMessageSummary(parsed)
+	if blockCount <= 0 {
+		blockCount = 1
+	}
+
+	asciiChars := 0
+	nonASCIIChars := 0
+	for _, r := range summary {
+		if r <= 127 {
+			asciiChars++
+			continue
+		}
+		nonASCIIChars++
+	}
+
+	lexicalTokens := nonASCIIChars
+	if asciiChars > 0 {
+		lexicalTokens += (asciiChars + 3) / 4
+	}
+	wordCount := len(strings.Fields(summary))
+	if wordCount > lexicalTokens {
+		lexicalTokens = wordCount
+	}
+	if lexicalTokens == 0 {
+		lexicalTokens = 1
+	}
+
+	structuralTokens := claudeMaxSimBaseOverheadTokens + blockCount*claudeMaxSimPerBlockOverhead
+	rawInputTokens := structuralTokens + lexicalTokens
+
+	maxInputTokens := clampInt(totalWindowTokens/claudeMaxSimContextDivisor, claudeMaxSimInputMinTokens, claudeMaxSimInputMaxTokens)
+	if totalWindowTokens <= claudeMaxSimInputMinTokens+1 {
+		maxInputTokens = totalWindowTokens - 1
+	}
+	if maxInputTokens <= 0 {
+		return totalWindowTokens
+	}
+
+	minInputTokens := 1
+	if totalWindowTokens > claudeMaxSimInputMinTokens+1 {
+		minInputTokens = claudeMaxSimInputMinTokens
+	}
+	return clampInt(rawInputTokens, minInputTokens, maxInputTokens)
+}
+
+func extractTailUserMessageSummary(parsed *ParsedRequest) (string, int) {
+	if parsed == nil || len(parsed.Messages) == 0 {
+		return "", 1
+	}
+	for i := len(parsed.Messages) - 1; i >= 0; i-- {
+		message, ok := parsed.Messages[i].(map[string]any)
+		if !ok {
+			continue
+		}
+		role, _ := message["role"].(string)
+		if !strings.EqualFold(strings.TrimSpace(role), "user") {
+			continue
+		}
+		summary, blockCount := summarizeUserContentBlocks(message["content"])
+		if blockCount <= 0 {
+			blockCount = 1
+		}
+		return summary, blockCount
+	}
+	return "", 1
+}
+
+func summarizeUserContentBlocks(content any) (string, int) {
+	appendSegment := func(segments []string, raw string) []string {
+		normalized := strings.Join(strings.Fields(strings.TrimSpace(raw)), " ")
+		if normalized == "" {
+			return segments
+		}
+		return append(segments, normalized)
+	}
+
+	switch value := content.(type) {
+	case string:
+		return trimClaudeMaxSummary(value), 1
+	case []any:
+		if len(value) == 0 {
+			return "", 1
+		}
+		segments := make([]string, 0, len(value))
+		for _, blockRaw := range value {
+			block, ok := blockRaw.(map[string]any)
+			if !ok {
+				continue
+			}
+			blockType, _ := block["type"].(string)
+			switch blockType {
+			case "text":
+				if text, ok := block["text"].(string); ok {
+					segments = appendSegment(segments, text)
+				}
+			case "tool_result":
+				nestedSummary, _ := summarizeUserContentBlocks(block["content"])
+				segments = appendSegment(segments, nestedSummary)
+			case "tool_use":
+				if name, ok := block["name"].(string); ok {
+					segments = appendSegment(segments, name)
+				}
+			default:
+				if text, ok := block["text"].(string); ok {
+					segments = appendSegment(segments, text)
+				}
+			}
+		}
+		return trimClaudeMaxSummary(strings.Join(segments, " ")), len(value)
+	default:
+		return "", 1
+	}
+}
+
+func trimClaudeMaxSummary(summary string) string {
+	normalized := strings.Join(strings.Fields(strings.TrimSpace(summary)), " ")
+	if normalized == "" {
+		return ""
+	}
+	runes := []rune(normalized)
+	if len(runes) > claudeMaxSimSummaryMaxRunes {
+		return string(runes[:claudeMaxSimSummaryMaxRunes])
+	}
+	return normalized
+}
+
+func clampInt(v, minValue, maxValue int) int {
+	if minValue > maxValue {
+		return minValue
+	}
+	if v < minValue {
+		return minValue
+	}
+	if v > maxValue {
+		return maxValue
+	}
+	return v
+}
+
 // RecordUsageInput 记录使用量的输入参数
 type RecordUsageInput struct {
 	Result            *ForwardResult
+	ParsedRequest     *ParsedRequest
 	APIKey            *APIKey
 	User              *User
 	Account           *Account
@@ -5601,9 +5829,25 @@ func (s *GatewayService) RecordUsage(ctx context.Context, input *RecordUsageInpu
 		result.Usage.InputTokens = 0
 	}
 
+	// Claude 分组模拟：将无写缓存 usage 映射为 claude-max 风格的 1h cache creation。
+	simulatedClaudeMax := false
+	if shouldSimulateClaudeMaxUsage(input) {
+		beforeInputTokens := result.Usage.InputTokens
+		simulatedClaudeMax = applyClaudeMaxUsageSimulation(result, input.ParsedRequest)
+		if simulatedClaudeMax {
+			logger.LegacyPrintf("service.gateway", "simulate_claude_max_usage: model=%s account=%d input_tokens:%d->%d cache_creation_1h=%d",
+				result.Model,
+				account.ID,
+				beforeInputTokens,
+				result.Usage.InputTokens,
+				result.Usage.CacheCreation1hTokens,
+			)
+		}
+	}
+
 	// Cache TTL Override: 确保计费时 token 分类与账号设置一致
 	cacheTTLOverridden := false
-	if account.IsCacheTTLOverrideEnabled() {
+	if account.IsCacheTTLOverrideEnabled() && !simulatedClaudeMax {
 		applyCacheTTLOverride(&result.Usage, account.GetCacheTTLOverrideTarget())
 		cacheTTLOverridden = (result.Usage.CacheCreation5mTokens + result.Usage.CacheCreation1hTokens) > 0
 	}

backend/internal/service/group.go

@@ -47,6 +47,9 @@ type Group struct {
 	// MCP XML 协议注入开关（仅 antigravity 平台使用）
 	MCPXMLInject bool
 
+	// Claude usage 模拟开关：将无写缓存 usage 模拟为 claude-max 风格
+	SimulateClaudeMaxEnabled bool
+
 	// 支持的模型系列（仅 antigravity 平台使用）
 	// 可选值: claude, gemini_text, gemini_image
 	SupportedModelScopes []string