Merge pull request #3916 from Wei-Shaw/fix/reasoning-effort-model-candidates

fix: effort 提取改用模型候选列表，修复后缀模型用量元数据丢失

backend/internal/service/openai_gateway_chat_completions_raw.go

@@ -78,7 +78,7 @@ func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
 	// 2. Resolve model mapping (same as ForwardAsChatCompletions)
 	billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
 	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
-	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, firstNonEmpty(upstreamModel, billingModel, originalModel))
+	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, upstreamModel, billingModel, originalModel)
 	// 国产模型默认 effort 补充：需要 mappedModel 判定，推迟到 billingModel 算出之后。
 	reasoningEffort = ApplyThinkingEnabledFallback(reasoningEffort, body, billingModel)
 

backend/internal/service/openai_gateway_forward.go

@@ -755,7 +755,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		}
 		defer func() { _ = resp.Body.Close() }()
 
-		reasoningEffort := extractOpenAIReasoningEffortFromBody(body, firstNonEmpty(upstreamModel, billingModel, originalModel))
+		reasoningEffort := extractOpenAIReasoningEffortFromBody(body, upstreamModel, billingModel, originalModel)
 		// 国产模型默认 effort 补充：此处 reqModel 已被 mapping 重写为 billingModel（见
 		// line 2510-2515 的 GetMappedModel + reqModel 赋值），可直接作为 mappedModel。
 		reasoningEffort = ApplyThinkingEnabledFallback(reasoningEffort, body, reqModel)

backend/internal/service/openai_gateway_messages_chat_fallback.go

@@ -72,7 +72,7 @@ func (s *OpenAIGatewayService) forwardAnthropicViaRawChatCompletions(
 		chatReq.StreamOptions = &apicompat.ChatStreamOptions{IncludeUsage: true}
 	}
 
-	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, firstNonEmpty(upstreamModel, billingModel, originalModel))
+	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, upstreamModel, billingModel, originalModel)
 	reasoningEffort = ApplyThinkingEnabledFallback(reasoningEffort, body, billingModel)
 	serviceTier := extractOpenAIServiceTierFromBody(body)
 

backend/internal/service/openai_gateway_request_body.go

@@ -331,6 +331,17 @@ func deriveOpenAIReasoningEffortFromModel(model string) string {
 	return normalizeOpenAIReasoningEffortForModel(parts[len(parts)-1], modelID)
 }
 
+// deriveOpenAIReasoningEffortFromModelCandidates 依次对每个候选模型做后缀推导，
+// 返回第一个非空结果。
+func deriveOpenAIReasoningEffortFromModelCandidates(models []string) string {
+	for _, model := range models {
+		if value := deriveOpenAIReasoningEffortFromModel(model); value != "" {
+			return value
+		}
+	}
+	return ""
+}
+
 type openAIRequestView struct {
 	body               []byte
 	Model              string
@@ -571,20 +582,24 @@ func detectOpenAIPassthroughInstructionsRejectReason(reqModel string, body []byt
 	return ""
 }
 
-func extractOpenAIReasoningEffortFromBody(body []byte, requestedModel string) *string {
+// extractOpenAIReasoningEffortFromBody 按优先级传入模型候选（如 upstreamModel,
+// billingModel, originalModel）：显式 effort 的模型归一化（max 保留判定）用第一个
+// 非空候选；body 未携带 effort 时的模型后缀推导依次尝试每个候选——OAuth 的
+// normalizeCodexModel 会剥掉 upstreamModel 的 effort 后缀，只有原始模型名还留着。
+func extractOpenAIReasoningEffortFromBody(body []byte, modelCandidates ...string) *string {
 	reasoningEffort := strings.TrimSpace(gjson.GetBytes(body, "reasoning.effort").String())
 	if reasoningEffort == "" {
 		reasoningEffort = strings.TrimSpace(gjson.GetBytes(body, "reasoning_effort").String())
 	}
 	if reasoningEffort != "" {
-		normalized := normalizeOpenAIReasoningEffortForModel(reasoningEffort, requestedModel)
+		normalized := normalizeOpenAIReasoningEffortForModel(reasoningEffort, firstNonEmpty(modelCandidates...))
 		if normalized == "" {
 			return nil
 		}
 		return &normalized
 	}
 
-	value := deriveOpenAIReasoningEffortFromModel(requestedModel)
+	value := deriveOpenAIReasoningEffortFromModelCandidates(modelCandidates)
 	if value == "" {
 		return nil
 	}
@@ -1159,15 +1174,16 @@ func getOpenAIRequestBodyMap(_ *gin.Context, body []byte) (map[string]any, error
 	return reqBody, nil
 }
 
-func extractOpenAIReasoningEffort(reqBody map[string]any, requestedModel string) *string {
-	if value, present := getOpenAIReasoningEffortFromReqBody(reqBody, requestedModel); present {
+// extractOpenAIReasoningEffort 的模型候选语义同 extractOpenAIReasoningEffortFromBody。
+func extractOpenAIReasoningEffort(reqBody map[string]any, modelCandidates ...string) *string {
+	if value, present := getOpenAIReasoningEffortFromReqBody(reqBody, firstNonEmpty(modelCandidates...)); present {
 		if value == "" {
 			return nil
 		}
 		return &value
 	}
 
-	value := deriveOpenAIReasoningEffortFromModel(requestedModel)
+	value := deriveOpenAIReasoningEffortFromModelCandidates(modelCandidates)
 	if value == "" {
 		return nil
 	}

backend/internal/service/openai_gateway_responses_chat_fallback.go

@@ -48,7 +48,7 @@ func (s *OpenAIGatewayService) forwardResponsesViaRawChatCompletions(
 
 	billingModel := resolveOpenAIForwardModel(account, originalModel, "")
 	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
-	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, firstNonEmpty(upstreamModel, billingModel, originalModel))
+	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, upstreamModel, billingModel, originalModel)
 	// 国产模型默认 effort 补充：需要 mappedModel 判定，推迟到 billingModel 算出之后。
 	reasoningEffort = ApplyThinkingEnabledFallback(reasoningEffort, body, billingModel)
 	chatReq.Model = upstreamModel

backend/internal/service/openai_reasoning_effort_candidates_test.go

@@ -0,0 +1,122 @@
+package service
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/Wei-Shaw/sub2api/internal/config"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+	"github.com/tidwall/gjson"
+)
+
+func TestExtractOpenAIReasoningEffortFromBodyModelCandidates(t *testing.T) {
+	bodyWithoutEffort := []byte(`{"model":"whatever","input":"hello"}`)
+	bodyWithMax := []byte(`{"model":"sol","reasoning":{"effort":"max"},"input":"hello"}`)
+
+	tests := []struct {
+		name       string
+		body       []byte
+		candidates []string
+		want       string // "" 表示期望 nil
+	}{
+		{
+			name:       "后缀推导回退到原始模型（OAuth 上游模型已剥后缀）",
+			body:       bodyWithoutEffort,
+			candidates: []string{"gpt-5.4", "gpt-5.4", "gpt-5.4-xhigh"},
+			want:       "xhigh",
+		},
+		{
+			name:       "GPT-5.6 后缀 max 经原始模型推导保留",
+			body:       bodyWithoutEffort,
+			candidates: []string{"gpt-5.6-sol", "gpt-5.6-sol", "gpt-5.6-sol-max"},
+			want:       "max",
+		},
+		{
+			name:       "显式 max 用第一个非空候选（映射后模型）判定",
+			body:       bodyWithMax,
+			candidates: []string{"gpt-5.6-sol", "sol"},
+			want:       "max",
+		},
+		{
+			name:       "显式 max 非 5.6 首候选仍折叠为 xhigh",
+			body:       bodyWithMax,
+			candidates: []string{"gpt-5.4", "sol"},
+			want:       "xhigh",
+		},
+		{
+			name:       "所有候选均无后缀时返回 nil",
+			body:       bodyWithoutEffort,
+			candidates: []string{"gpt-5.4", "gpt-5.4", "gpt-5.4"},
+			want:       "",
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			got := extractOpenAIReasoningEffortFromBody(tt.body, tt.candidates...)
+			if tt.want == "" {
+				require.Nil(t, got)
+				return
+			}
+			require.NotNil(t, got)
+			require.Equal(t, tt.want, *got)
+		})
+	}
+}
+
+func TestExtractOpenAIReasoningEffortModelCandidates(t *testing.T) {
+	reqBody := map[string]any{"model": "gpt-5.3-codex-high", "input": "hello"}
+
+	got := extractOpenAIReasoningEffort(reqBody, "gpt-5.3-codex", "gpt-5.3-codex-high")
+
+	require.NotNil(t, got)
+	require.Equal(t, "high", *got)
+}
+
+// 回归：OAuth 账号请求后缀式模型（无显式 reasoning 字段）时，上游模型被
+// normalizeCodexModel 剥掉 effort 后缀，用量元数据的 effort 必须仍能从
+// 原始模型名后缀推导出来。
+func TestOpenAIGatewayServiceForwardOAuthDerivesEffortFromSuffixModel(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+	upstream := &httpUpstreamRecorder{
+		resp: &http.Response{
+			StatusCode: http.StatusOK,
+			Header:     http.Header{"Content-Type": []string{"application/json"}},
+			Body:       io.NopCloser(strings.NewReader(`{"usage":{"input_tokens":1,"output_tokens":2}}`)),
+		},
+	}
+	cfg := &config.Config{}
+	cfg.Security.URLAllowlist.Enabled = false
+	svc := &OpenAIGatewayService{cfg: cfg, httpUpstream: upstream}
+	account := &Account{
+		ID:          11,
+		Name:        "openai-oauth-suffix",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeOAuth,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"access_token":       "oauth-token",
+			"chatgpt_account_id": "chatgpt-acc",
+		},
+		Status:      StatusActive,
+		Schedulable: true,
+	}
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	c.Request = httptest.NewRequest(http.MethodPost, "/openai/v1/responses", nil)
+	SetOpenAIClientTransport(c, OpenAIClientTransportHTTP)
+
+	body := []byte(`{"model":"gpt-5.3-codex-xhigh","instructions":"suffix-test","input":"hello","stream":false}`)
+	result, err := svc.Forward(context.Background(), c, account, body)
+
+	require.NoError(t, err)
+	require.NotNil(t, result)
+	require.Equal(t, "gpt-5.3-codex", gjson.GetBytes(upstream.lastBody, "model").String())
+	require.NotNil(t, result.ReasoningEffort)
+	require.Equal(t, "xhigh", *result.ReasoningEffort)
+}

backend/internal/service/openai_ws_forwarder_ingress.go

@@ -920,7 +920,7 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 					Model:           originalModel,
 					UpstreamModel:   mappedModel,
 					ServiceTier:     extractOpenAIServiceTierFromBody(payload),
-					ReasoningEffort: ApplyThinkingEnabledFallback(extractOpenAIReasoningEffortFromBody(payload, firstNonEmpty(mappedModel, originalModel)), payload, mappedModel),
+					ReasoningEffort: ApplyThinkingEnabledFallback(extractOpenAIReasoningEffortFromBody(payload, mappedModel, originalModel), payload, mappedModel),
 					Stream:          reqStream,
 					OpenAIWSMode:    true,
 					ResponseHeaders: lease.HandshakeHeaders(),

backend/internal/service/openai_ws_forwarder_v2.go

@@ -693,7 +693,7 @@ func (s *OpenAIGatewayService) forwardOpenAIWSV2(
 		ImageCount:       imageCounter.Count(),
 		ImageOutputSizes: imageCounter.Sizes(),
 		ServiceTier:      extractOpenAIServiceTier(reqBody),
-		ReasoningEffort:  extractOpenAIReasoningEffort(reqBody, firstNonEmpty(mappedModel, originalModel)),
+		ReasoningEffort:  extractOpenAIReasoningEffort(reqBody, mappedModel, originalModel),
 		Stream:           reqStream,
 		OpenAIWSMode:     true,
 		ResponseHeaders:  lease.HandshakeHeaders(),

backend/internal/service/openai_ws_http_bridge.go

@@ -263,7 +263,7 @@ func (s *OpenAIGatewayService) proxyOpenAIWSHTTPBridgeTurn(
 			Model:           originalModel,
 			UpstreamModel:   mappedModel,
 			ServiceTier:     extractOpenAIServiceTierFromBody(body),
-			ReasoningEffort: ApplyThinkingEnabledFallback(extractOpenAIReasoningEffortFromBody(body, firstNonEmpty(mappedModel, originalModel)), body, mappedModel),
+			ReasoningEffort: ApplyThinkingEnabledFallback(extractOpenAIReasoningEffortFromBody(body, mappedModel, originalModel), body, mappedModel),
 			Stream:          reqStream,
 			OpenAIWSMode:    true,
 			ResponseHeaders: cloneHeader(resp.Header),