fix: gpt->claude格式转换对齐effort映射和fast

backend/internal/pkg/apicompat/anthropic_responses_test.go

@@ -631,7 +631,8 @@ func TestAnthropicToResponses_ThinkingEnabled(t *testing.T) {
 	resp, err := AnthropicToResponses(req)
 	require.NoError(t, err)
 	require.NotNil(t, resp.Reasoning)
-	assert.Equal(t, "high", resp.Reasoning.Effort)
+	// thinking.type is ignored for effort; default xhigh applies.
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
 	assert.Equal(t, "auto", resp.Reasoning.Summary)
 	assert.Contains(t, resp.Include, "reasoning.encrypted_content")
 	assert.NotContains(t, resp.Include, "reasoning.summary")
@@ -648,7 +649,8 @@ func TestAnthropicToResponses_ThinkingAdaptive(t *testing.T) {
 	resp, err := AnthropicToResponses(req)
 	require.NoError(t, err)
 	require.NotNil(t, resp.Reasoning)
-	assert.Equal(t, "medium", resp.Reasoning.Effort)
+	// thinking.type is ignored for effort; default xhigh applies.
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
 	assert.Equal(t, "auto", resp.Reasoning.Summary)
 	assert.NotContains(t, resp.Include, "reasoning.summary")
 }
@@ -663,8 +665,9 @@ func TestAnthropicToResponses_ThinkingDisabled(t *testing.T) {
 
 	resp, err := AnthropicToResponses(req)
 	require.NoError(t, err)
-	assert.Nil(t, resp.Reasoning)
-	assert.NotContains(t, resp.Include, "reasoning.summary")
+	// Default effort applies (high → xhigh) even when thinking is disabled.
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
 }
 
 func TestAnthropicToResponses_NoThinking(t *testing.T) {
@@ -676,7 +679,93 @@ func TestAnthropicToResponses_NoThinking(t *testing.T) {
 
 	resp, err := AnthropicToResponses(req)
 	require.NoError(t, err)
-	assert.Nil(t, resp.Reasoning)
+	// Default effort applies (high → xhigh) when no thinking/output_config is set.
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
+}
+
+// ---------------------------------------------------------------------------
+// output_config.effort override tests
+// ---------------------------------------------------------------------------
+
+func TestAnthropicToResponses_OutputConfigOverridesDefault(t *testing.T) {
+	// Default is xhigh, but output_config.effort="low" overrides. low→low after mapping.
+	req := &AnthropicRequest{
+		Model:        "gpt-5.2",
+		MaxTokens:    1024,
+		Messages:     []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		Thinking:     &AnthropicThinking{Type: "enabled", BudgetTokens: 10000},
+		OutputConfig: &AnthropicOutputConfig{Effort: "low"},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "low", resp.Reasoning.Effort)
+	assert.Equal(t, "auto", resp.Reasoning.Summary)
+}
+
+func TestAnthropicToResponses_OutputConfigWithoutThinking(t *testing.T) {
+	// No thinking field, but output_config.effort="medium" → creates reasoning.
+	// medium→high after mapping.
+	req := &AnthropicRequest{
+		Model:        "gpt-5.2",
+		MaxTokens:    1024,
+		Messages:     []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		OutputConfig: &AnthropicOutputConfig{Effort: "medium"},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "high", resp.Reasoning.Effort)
+	assert.Equal(t, "auto", resp.Reasoning.Summary)
+}
+
+func TestAnthropicToResponses_OutputConfigHigh(t *testing.T) {
+	// output_config.effort="high" → mapped to "xhigh".
+	req := &AnthropicRequest{
+		Model:        "gpt-5.2",
+		MaxTokens:    1024,
+		Messages:     []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		OutputConfig: &AnthropicOutputConfig{Effort: "high"},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
+	assert.Equal(t, "auto", resp.Reasoning.Summary)
+}
+
+func TestAnthropicToResponses_NoOutputConfig(t *testing.T) {
+	// No output_config → default xhigh regardless of thinking.type.
+	req := &AnthropicRequest{
+		Model:     "gpt-5.2",
+		MaxTokens: 1024,
+		Messages:  []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		Thinking:  &AnthropicThinking{Type: "enabled", BudgetTokens: 10000},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
+}
+
+func TestAnthropicToResponses_OutputConfigWithoutEffort(t *testing.T) {
+	// output_config present but effort empty (e.g. only format set) → default xhigh.
+	req := &AnthropicRequest{
+		Model:        "gpt-5.2",
+		MaxTokens:    1024,
+		Messages:     []AnthropicMessage{{Role: "user", Content: json.RawMessage(`"Hello"`)}},
+		OutputConfig: &AnthropicOutputConfig{},
+	}
+
+	resp, err := AnthropicToResponses(req)
+	require.NoError(t, err)
+	require.NotNil(t, resp.Reasoning)
+	assert.Equal(t, "xhigh", resp.Reasoning.Effort)
 }
 
 // ---------------------------------------------------------------------------

backend/internal/pkg/apicompat/anthropic_to_responses.go

@@ -45,18 +45,16 @@ func AnthropicToResponses(req *AnthropicRequest) (*ResponsesRequest, error) {
 		out.Tools = convertAnthropicToolsToResponses(req.Tools)
 	}
 
-	// Convert thinking → reasoning.
-	// generate_summary="auto" causes the upstream to emit reasoning_summary_text
-	// streaming events; the include array only needs reasoning.encrypted_content
-	// (already set above) for content continuity.
-	if req.Thinking != nil {
-		switch req.Thinking.Type {
-		case "enabled":
-			out.Reasoning = &ResponsesReasoning{Effort: "high", Summary: "auto"}
-		case "adaptive":
-			out.Reasoning = &ResponsesReasoning{Effort: "medium", Summary: "auto"}
-		}
-		// "disabled" or unknown → omit reasoning
+	// Determine reasoning effort: only output_config.effort controls the
+	// level; thinking.type is ignored. Default is xhigh when unset.
+	// Anthropic levels map to OpenAI: low→low, medium→high, high→xhigh.
+	effort := "high" // default → maps to xhigh
+	if req.OutputConfig != nil && req.OutputConfig.Effort != "" {
+		effort = req.OutputConfig.Effort
+	}
+	out.Reasoning = &ResponsesReasoning{
+		Effort:  mapAnthropicEffortToResponses(effort),
+		Summary: "auto",
 	}
 
 	// Convert tool_choice
@@ -380,6 +378,23 @@ func extractAnthropicTextFromBlocks(blocks []AnthropicContentBlock) string {
 	return strings.Join(parts, "\n\n")
 }
 
+// mapAnthropicEffortToResponses converts Anthropic reasoning effort levels to
+// OpenAI Responses API effort levels.
+//
+//	low    → low
+//	medium → high
+//	high   → xhigh
+func mapAnthropicEffortToResponses(effort string) string {
+	switch effort {
+	case "medium":
+		return "high"
+	case "high":
+		return "xhigh"
+	default:
+		return effort // "low" and any unknown values pass through unchanged
+	}
+}
+
 // convertAnthropicToolsToResponses maps Anthropic tool definitions to
 // Responses API tools. Server-side tools like web_search are mapped to their
 // OpenAI equivalents; regular tools become function tools.

backend/internal/pkg/apicompat/types.go

@@ -12,17 +12,23 @@ import "encoding/json"
 
 // AnthropicRequest is the request body for POST /v1/messages.
 type AnthropicRequest struct {
-	Model       string             `json:"model"`
-	MaxTokens   int                `json:"max_tokens"`
-	System      json.RawMessage    `json:"system,omitempty"` // string or []AnthropicContentBlock
-	Messages    []AnthropicMessage `json:"messages"`
-	Tools       []AnthropicTool    `json:"tools,omitempty"`
-	Stream      bool               `json:"stream,omitempty"`
-	Temperature *float64           `json:"temperature,omitempty"`
-	TopP        *float64           `json:"top_p,omitempty"`
-	StopSeqs    []string           `json:"stop_sequences,omitempty"`
-	Thinking    *AnthropicThinking `json:"thinking,omitempty"`
-	ToolChoice  json.RawMessage    `json:"tool_choice,omitempty"`
+	Model        string                 `json:"model"`
+	MaxTokens    int                    `json:"max_tokens"`
+	System       json.RawMessage        `json:"system,omitempty"` // string or []AnthropicContentBlock
+	Messages     []AnthropicMessage     `json:"messages"`
+	Tools        []AnthropicTool        `json:"tools,omitempty"`
+	Stream       bool                   `json:"stream,omitempty"`
+	Temperature  *float64               `json:"temperature,omitempty"`
+	TopP         *float64               `json:"top_p,omitempty"`
+	StopSeqs     []string               `json:"stop_sequences,omitempty"`
+	Thinking     *AnthropicThinking     `json:"thinking,omitempty"`
+	ToolChoice   json.RawMessage        `json:"tool_choice,omitempty"`
+	OutputConfig *AnthropicOutputConfig `json:"output_config,omitempty"`
+}
+
+// AnthropicOutputConfig controls output generation parameters.
+type AnthropicOutputConfig struct {
+	Effort string `json:"effort,omitempty"` // "low" | "medium" | "high"
 }
 
 // AnthropicThinking configures extended thinking in the Anthropic API.
@@ -156,6 +162,7 @@ type ResponsesRequest struct {
 	Store           *bool               `json:"store,omitempty"`
 	Reasoning       *ResponsesReasoning `json:"reasoning,omitempty"`
 	ToolChoice      json.RawMessage     `json:"tool_choice,omitempty"`
+	ServiceTier     string              `json:"service_tier,omitempty"`
 }
 
 // ResponsesReasoning configures reasoning effort in the Responses API.

backend/internal/service/gateway_beta_test.go

@@ -148,6 +148,32 @@ func TestBuildBetaTokenSet(t *testing.T) {
 	require.Empty(t, empty)
 }
 
+func TestContainsBetaToken(t *testing.T) {
+	tests := []struct {
+		name   string
+		header string
+		token  string
+		want   bool
+	}{
+		{"present in middle", "oauth-2025-04-20,fast-mode-2026-02-01,interleaved-thinking-2025-05-14", "fast-mode-2026-02-01", true},
+		{"present at start", "fast-mode-2026-02-01,oauth-2025-04-20", "fast-mode-2026-02-01", true},
+		{"present at end", "oauth-2025-04-20,fast-mode-2026-02-01", "fast-mode-2026-02-01", true},
+		{"only token", "fast-mode-2026-02-01", "fast-mode-2026-02-01", true},
+		{"not present", "oauth-2025-04-20,interleaved-thinking-2025-05-14", "fast-mode-2026-02-01", false},
+		{"with spaces", "oauth-2025-04-20, fast-mode-2026-02-01 , interleaved-thinking-2025-05-14", "fast-mode-2026-02-01", true},
+		{"empty header", "", "fast-mode-2026-02-01", false},
+		{"empty token", "fast-mode-2026-02-01", "", false},
+		{"partial match", "fast-mode-2026-02-01-extra", "fast-mode-2026-02-01", false},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := containsBetaToken(tt.header, tt.token)
+			require.Equal(t, tt.want, got)
+		})
+	}
+}
+
 func TestStripBetaTokensWithSet_EmptyDropSet(t *testing.T) {
 	header := "oauth-2025-04-20,interleaved-thinking-2025-05-14"
 	got := stripBetaTokensWithSet(header, map[string]struct{}{})

backend/internal/service/gateway_service.go

@@ -5341,6 +5341,19 @@ func droppedBetaSet(extra ...string) map[string]struct{} {
 	return m
 }
 
+// containsBetaToken checks if a comma-separated header value contains the given token.
+func containsBetaToken(header, token string) bool {
+	if header == "" || token == "" {
+		return false
+	}
+	for _, p := range strings.Split(header, ",") {
+		if strings.TrimSpace(p) == token {
+			return true
+		}
+	}
+	return false
+}
+
 func buildBetaTokenSet(tokens []string) map[string]struct{} {
 	m := make(map[string]struct{}, len(tokens))
 	for _, t := range tokens {

backend/internal/service/openai_gateway_messages.go

@@ -12,6 +12,7 @@ import (
 	"time"
 
 	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
 	"github.com/gin-gonic/gin"
@@ -46,6 +47,11 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic(
 		return nil, fmt.Errorf("convert anthropic to responses: %w", err)
 	}
 
+	// 2b. Handle BetaFastMode → service_tier: "priority"
+	if containsBetaToken(c.GetHeader("anthropic-beta"), claude.BetaFastMode) {
+		responsesReq.ServiceTier = "priority"
+	}
+
 	// 3. Model mapping
 	mappedModel := account.GetMappedModel(originalModel)
 	// 分组级降级：账号未映射时使用分组默认映射模型
@@ -94,6 +100,12 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic(
 		return nil, fmt.Errorf("build upstream request: %w", err)
 	}
 
+	// Override session_id with a deterministic UUID derived from the sticky
+	// session key (buildUpstreamRequest may have set it to the raw value).
+	if promptCacheKey != "" {
+		upstreamReq.Header.Set("session_id", generateSessionUUID(promptCacheKey))
+	}
+
 	// 7. Send request
 	proxyURL := ""
 	if account.Proxy != nil {
@@ -160,6 +172,18 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic(
 		result, handleErr = s.handleAnthropicNonStreamingResponse(resp, c, originalModel, mappedModel, startTime)
 	}
 
+	// Propagate ServiceTier and ReasoningEffort to result for billing
+	if handleErr == nil && result != nil {
+		if responsesReq.ServiceTier != "" {
+			st := responsesReq.ServiceTier
+			result.ServiceTier = &st
+		}
+		if responsesReq.Reasoning != nil && responsesReq.Reasoning.Effort != "" {
+			re := responsesReq.Reasoning.Effort
+			result.ReasoningEffort = &re
+		}
+	}
+
 	// Extract and save Codex usage snapshot from response headers (for OAuth accounts)
 	if handleErr == nil && account.Type == AccountTypeOAuth {
 		if snapshot := ParseCodexRateLimitHeaders(resp.Header); snapshot != nil {