Merge pull request #2143 from alfadb/fix/openai-apikey-cc-default-routing

修复：APIKey 账户上游不支持 OpenAI Responses API 时的 Chat Completions 路由回退

backend/internal/service/account_test_service.go

@@ -21,6 +21,7 @@ import (
 	"github.com/Wei-Shaw/sub2api/internal/pkg/claude"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/geminicli"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai_compat"
 	"github.com/Wei-Shaw/sub2api/internal/util/urlvalidator"
 	"github.com/gin-gonic/gin"
 	"github.com/google/uuid"
@@ -554,7 +555,16 @@ func (s *AccountTestService) testOpenAIAccountConnection(c *gin.Context, account
 		if err != nil {
 			return s.sendErrorAndEnd(c, fmt.Sprintf("Invalid base URL: %s", err.Error()))
 		}
-		apiURL = strings.TrimSuffix(normalizedBaseURL, "/") + "/responses"
+		// 账号已被探测为不支持 Responses（如 DeepSeek/Kimi 等）时，丢出明确提示。
+		// 账号本身可用（网关会走 CC 直转），仅测试入口需要补齐 CC SSE 处理逻辑。
+		// TODO：实现 CC 格式的账号测试路径（需专门的 CC SSE handler）。
+		if !openai_compat.ShouldUseResponsesAPI(account.Extra) {
+			return s.sendErrorAndEnd(c,
+				"账号已被探测为不支持 OpenAI Responses API（如 DeepSeek/Kimi 等三方兼容上游），"+
+					"账号本身可正常使用，但当前测试接口仅支持 Responses API 路径。请直接通过实际 API 调用验证。",
+			)
+		}
+		apiURL = buildOpenAIResponsesURL(normalizedBaseURL)
 	} else {
 		return s.sendErrorAndEnd(c, fmt.Sprintf("Unsupported account type: %s", account.Type))
 	}

backend/internal/service/openai_apikey_responses_probe.go

@@ -0,0 +1,149 @@
+package service
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai_compat"
+)
+
+// openaiResponsesProbeTimeout 是探测请求的超时时长。
+// 探测必须快速失败——超时不应阻塞账号创建/更新流程。
+const openaiResponsesProbeTimeout = 8 * time.Second
+
+// openaiResponsesProbePayload 是探测使用的最小 Responses 请求体。
+// 仅作能力探测，不期望响应内容质量；Stream=false 减少 SSE 解析开销。
+//
+// 注意：探测的目标是区分"端点存在"与"端点不存在"——只要上游返回非 404 的
+// 4xx/5xx（如 400 invalid_request_error / 401 unauthorized / 422 等），
+// 都视为"端点存在 → 支持 Responses"。仅 404 / 405 视为"端点不存在"。
+func openaiResponsesProbePayload(modelID string) []byte {
+	if strings.TrimSpace(modelID) == "" {
+		modelID = openai.DefaultTestModel
+	}
+	body, _ := json.Marshal(map[string]any{
+		"model": modelID,
+		"input": []map[string]any{
+			{
+				"role": "user",
+				"content": []map[string]any{
+					{"type": "input_text", "text": "hi"},
+				},
+			},
+		},
+		"instructions": openai.DefaultInstructions,
+		"stream":       false,
+	})
+	return body
+}
+
+// ProbeOpenAIAPIKeyResponsesSupport 探测 OpenAI APIKey 账号上游是否支持
+// /v1/responses 端点，并将结果持久化到 accounts.extra.openai_responses_supported。
+//
+// 调用时机：账号创建/更新后，且仅当 platform=openai && type=apikey 时。
+//
+// 探测策略（参见包文档 internal/pkg/openai_compat）：
+//   - 上游 404 / 405 → 不支持，写 false
+//   - 上游 2xx / 其他 4xx（401/422/400 等）/ 5xx → 支持，写 true
+//   - 网络层失败（连接错误、超时）→ 不写标记，保持 unknown
+//     （后续请求仍按"现状即证据"默认走 Responses）
+//
+// 该方法是幂等的：重复调用会以最新探测结果覆盖标记。
+//
+// 关于失败处理：探测本身的失败不应阻塞账号创建——账号能创建/更新成功就够了，
+// 探测结果只影响后续路由优化。所有错误都仅记录日志，不向调用方传播。
+func (s *AccountTestService) ProbeOpenAIAPIKeyResponsesSupport(ctx context.Context, accountID int64) {
+	account, err := s.accountRepo.GetByID(ctx, accountID)
+	if err != nil {
+		logger.LegacyPrintf("service.openai_probe", "probe_load_account_failed: account_id=%d err=%v", accountID, err)
+		return
+	}
+	if account.Platform != PlatformOpenAI || account.Type != AccountTypeAPIKey {
+		// 仅 OpenAI APIKey 账号需要探测；其他账号类型无能力差异。
+		return
+	}
+
+	apiKey := account.GetOpenAIApiKey()
+	if apiKey == "" {
+		logger.LegacyPrintf("service.openai_probe", "probe_skip_no_apikey: account_id=%d", accountID)
+		return
+	}
+	baseURL := account.GetOpenAIBaseURL()
+	if baseURL == "" {
+		baseURL = "https://api.openai.com"
+	}
+	normalizedBaseURL, err := s.validateUpstreamBaseURL(baseURL)
+	if err != nil {
+		logger.LegacyPrintf("service.openai_probe", "probe_invalid_baseurl: account_id=%d base_url=%q err=%v", accountID, baseURL, err)
+		return
+	}
+
+	probeURL := buildOpenAIResponsesURL(normalizedBaseURL)
+
+	probeCtx, cancel := context.WithTimeout(ctx, openaiResponsesProbeTimeout)
+	defer cancel()
+
+	req, err := http.NewRequestWithContext(probeCtx, http.MethodPost, probeURL, bytes.NewReader(openaiResponsesProbePayload("")))
+	if err != nil {
+		logger.LegacyPrintf("service.openai_probe", "probe_build_request_failed: account_id=%d err=%v", accountID, err)
+		return
+	}
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer "+apiKey)
+	req.Header.Set("Accept", "application/json")
+
+	proxyURL := ""
+	if account.ProxyID != nil && account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+
+	resp, err := s.httpUpstream.DoWithTLS(req, proxyURL, account.ID, account.Concurrency, s.tlsFPProfileService.ResolveTLSProfile(account))
+	if err != nil {
+		// 网络层失败：不写标记，保持 unknown，下次重试或由网关 fallback 处理
+		logger.LegacyPrintf("service.openai_probe", "probe_request_failed: account_id=%d url=%s err=%v", accountID, probeURL, err)
+		return
+	}
+	defer func() {
+		_, _ = io.Copy(io.Discard, io.LimitReader(resp.Body, 1<<20))
+		_ = resp.Body.Close()
+	}()
+
+	supported := isResponsesEndpointSupportedByStatus(resp.StatusCode)
+
+	if err := s.accountRepo.UpdateExtra(ctx, accountID, map[string]any{
+		openai_compat.ExtraKeyResponsesSupported: supported,
+	}); err != nil {
+		logger.LegacyPrintf("service.openai_probe", "probe_persist_failed: account_id=%d supported=%v err=%v", accountID, supported, err)
+		return
+	}
+
+	logger.LegacyPrintf("service.openai_probe",
+		"probe_done: account_id=%d base_url=%s status=%d supported=%v",
+		accountID, normalizedBaseURL, resp.StatusCode, supported,
+	)
+}
+
+// isResponsesEndpointSupportedByStatus 根据探测响应的 HTTP 状态码判定上游
+// 是否暴露 /v1/responses 端点。
+//
+// 关键观察：第三方 OpenAI 兼容上游（DeepSeek/Kimi 等）对未知端点统一返回 404
+// 或 405；而 OpenAI 官方/有 Responses 实现的上游会因为请求体最简（缺字段）
+// 返回 400/422 等业务错误，但端点本身存在。
+//
+// 因此：仅 404 和 405 视为"端点不存在"，其他 status 视为"端点存在"。
+//
+// 5xx 也视为"端点存在"——上游偶发故障不应误判为不支持。
+func isResponsesEndpointSupportedByStatus(status int) bool {
+	switch status {
+	case http.StatusNotFound, http.StatusMethodNotAllowed:
+		return false
+	}
+	return true
+}

backend/internal/service/openai_gateway_chat_completions.go

@@ -15,6 +15,7 @@ import (
 
 	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
 	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/openai_compat"
 	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
 	"github.com/gin-gonic/gin"
 	"github.com/tidwall/gjson"
@@ -40,9 +41,18 @@ var cursorResponsesUnsupportedFields = []string{
 
 // ForwardAsChatCompletions accepts a Chat Completions request body, converts it
 // to OpenAI Responses API format, forwards to the OpenAI upstream, and converts
-// the response back to Chat Completions format. All account types (OAuth and API
-// Key) go through the Responses API conversion path since the upstream only
-// exposes the /v1/responses endpoint.
+// the response back to Chat Completions format.
+//
+// 历史背景：该函数原本对所有 OpenAI 账号无差别走 CC→Responses 转换 + /v1/responses
+// 端点——这在 OAuth（ChatGPT 内部 API 仅支持 Responses）和官方 APIKey 账号上是
+// 正确的，但 sub2api 接入 DeepSeek/Kimi/GLM 等第三方 OpenAI 兼容上游后假设破裂：
+// 这些上游普遍只支持 /v1/chat/completions，无 /v1/responses 端点。
+//
+// 当前路由策略（基于账号探测标记，详见 openai_compat.ShouldUseResponsesAPI）：
+//   - APIKey 账号 + 探测确认不支持 Responses → 走 forwardAsRawChatCompletions
+//     直转上游 /v1/chat/completions，不做协议转换
+//   - 其他所有情况（OAuth、APIKey 探测确认支持、未探测）→ 走原有 CC→Responses
+//     转换路径（保留旧行为，存量未探测账号零兼容破坏）
 func (s *OpenAIGatewayService) ForwardAsChatCompletions(
 	ctx context.Context,
 	c *gin.Context,
@@ -51,6 +61,12 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions(
 	promptCacheKey string,
 	defaultMappedModel string,
 ) (*OpenAIForwardResult, error) {
+	// 入口分流：APIKey 账号 + 已探测且确认上游不支持 Responses，走 CC 直转。
+	// 标记缺失（未探测）按"现状即证据"原则继续走下方原 Responses 转换路径。
+	if account.Type == AccountTypeAPIKey && !openai_compat.ShouldUseResponsesAPI(account.Extra) {
+		return s.forwardAsRawChatCompletions(ctx, c, account, body, defaultMappedModel)
+	}
+
 	startTime := time.Now()
 
 	// 1. Parse Chat Completions request

backend/internal/service/openai_gateway_chat_completions_raw.go

@@ -0,0 +1,399 @@
+package service
+
+import (
+	"bufio"
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/apicompat"
+	"github.com/Wei-Shaw/sub2api/internal/pkg/logger"
+	"github.com/Wei-Shaw/sub2api/internal/util/responseheaders"
+	"github.com/gin-gonic/gin"
+	"github.com/tidwall/gjson"
+	"go.uber.org/zap"
+)
+
+// openaiCCRawAllowedHeaders 是 CC 直转路径专用的客户端 header 透传白名单。
+//
+// **关键**：不能复用 openaiAllowedHeaders——后者含 Codex 客户端专属 header
+// （originator / session_id / x-codex-turn-state / x-codex-turn-metadata / conversation_id），
+// 这些在 ChatGPT OAuth 上游是必需的，但透传给 DeepSeek/Kimi/GLM 等第三方
+// OpenAI 兼容上游会造成：
+//   - 完全忽略（多数友好厂商）——隐性污染上游统计
+//   - 400 "unknown parameter"（严格上游）——可见错误
+//
+// 这里仅放行通用 HTTP header；content-type / authorization / accept 由上下文
+// 显式设置，不依赖透传。
+//
+// 参见决策记录：
+// pensieve/short-term/maxims/dont-reuse-shared-headers-whitelist-across-different-upstream-trust-domains
+var openaiCCRawAllowedHeaders = map[string]bool{
+	"accept-language": true,
+	"user-agent":      true,
+}
+
+// forwardAsRawChatCompletions 直转客户端的 Chat Completions 请求到上游
+// `{base_url}/v1/chat/completions`，**不**做 CC↔Responses 协议转换。
+//
+// 适用场景：account.platform=openai && account.type=apikey && 上游已被探测确认
+// 不支持 /v1/responses 端点（如 DeepSeek/Kimi/GLM/Qwen 等第三方 OpenAI 兼容上游）。
+//
+// 与 ForwardAsChatCompletions 的关键差异：
+//
+//   - 不调用 apicompat.ChatCompletionsToResponses，body 仅做模型 ID 改写
+//   - 上游 URL 拼到 /v1/chat/completions 而非 /v1/responses
+//   - 流式响应 SSE 直接透传给客户端（上游 chunk 已是 CC 格式）
+//   - 非流式响应 JSON 直接透传，仅按需提取 usage
+//   - 不应用 codex OAuth transform（APIKey 路径无 OAuth）
+//   - 不注入 prompt_cache_key（OAuth 专属机制）
+//
+// 调用入口：openai_gateway_chat_completions.go::ForwardAsChatCompletions
+// 在函数顶部按 openai_compat.ShouldUseResponsesAPI 分流。
+func (s *OpenAIGatewayService) forwardAsRawChatCompletions(
+	ctx context.Context,
+	c *gin.Context,
+	account *Account,
+	body []byte,
+	defaultMappedModel string,
+) (*OpenAIForwardResult, error) {
+	startTime := time.Now()
+
+	// 1. Parse minimal fields needed for routing/billing
+	originalModel := gjson.GetBytes(body, "model").String()
+	if originalModel == "" {
+		writeChatCompletionsError(c, http.StatusBadRequest, "invalid_request_error", "model is required")
+		return nil, fmt.Errorf("missing model in request")
+	}
+	clientStream := gjson.GetBytes(body, "stream").Bool()
+
+	// 1b. Extract reasoning effort and service tier from the raw body before any transformation.
+	reasoningEffort := extractOpenAIReasoningEffortFromBody(body, originalModel)
+	serviceTier := extractOpenAIServiceTierFromBody(body)
+
+	// 2. Resolve model mapping (same as ForwardAsChatCompletions)
+	billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel)
+	upstreamModel := normalizeOpenAIModelForUpstream(account, billingModel)
+
+	// 3. Rewrite model in body (no protocol conversion)
+	upstreamBody := body
+	if upstreamModel != originalModel {
+		upstreamBody = ReplaceModelInBody(body, upstreamModel)
+	}
+
+	// 4. Apply OpenAI fast policy on the CC body
+	updatedBody, policyErr := s.applyOpenAIFastPolicyToBody(ctx, account, upstreamModel, upstreamBody)
+	if policyErr != nil {
+		var blocked *OpenAIFastBlockedError
+		if errors.As(policyErr, &blocked) {
+			writeChatCompletionsError(c, http.StatusForbidden, "permission_error", blocked.Message)
+		}
+		return nil, policyErr
+	}
+	upstreamBody = updatedBody
+
+	logger.L().Debug("openai chat_completions raw: forwarding without protocol conversion",
+		zap.Int64("account_id", account.ID),
+		zap.String("original_model", originalModel),
+		zap.String("billing_model", billingModel),
+		zap.String("upstream_model", upstreamModel),
+		zap.Bool("stream", clientStream),
+	)
+
+	// 5. Build upstream request
+	apiKey := account.GetOpenAIApiKey()
+	if apiKey == "" {
+		return nil, fmt.Errorf("account %d missing api_key", account.ID)
+	}
+	baseURL := account.GetOpenAIBaseURL()
+	if baseURL == "" {
+		baseURL = "https://api.openai.com"
+	}
+	validatedURL, err := s.validateUpstreamBaseURL(baseURL)
+	if err != nil {
+		return nil, fmt.Errorf("invalid base_url: %w", err)
+	}
+	targetURL := buildOpenAIChatCompletionsURL(validatedURL)
+
+	upstreamReq, err := http.NewRequestWithContext(ctx, http.MethodPost, targetURL, bytes.NewReader(upstreamBody))
+	if err != nil {
+		return nil, fmt.Errorf("build upstream request: %w", err)
+	}
+	upstreamReq.Header.Set("Content-Type", "application/json")
+	upstreamReq.Header.Set("Authorization", "Bearer "+apiKey)
+	if clientStream {
+		upstreamReq.Header.Set("Accept", "text/event-stream")
+	} else {
+		upstreamReq.Header.Set("Accept", "application/json")
+	}
+
+	// 透传白名单中的客户端 header。详见 openaiCCRawAllowedHeaders 的设计说明。
+	for key, values := range c.Request.Header {
+		lowerKey := strings.ToLower(key)
+		if openaiCCRawAllowedHeaders[lowerKey] {
+			for _, v := range values {
+				upstreamReq.Header.Add(key, v)
+			}
+		}
+	}
+	customUA := account.GetOpenAIUserAgent()
+	if customUA != "" {
+		upstreamReq.Header.Set("user-agent", customUA)
+	}
+
+	// 6. Send request
+	proxyURL := ""
+	if account.Proxy != nil {
+		proxyURL = account.Proxy.URL()
+	}
+	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
+	if err != nil {
+		safeErr := sanitizeUpstreamErrorMessage(err.Error())
+		setOpsUpstreamError(c, 0, safeErr, "")
+		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+			Platform:           account.Platform,
+			AccountID:          account.ID,
+			AccountName:        account.Name,
+			UpstreamStatusCode: 0,
+			Kind:               "request_error",
+			Message:            safeErr,
+		})
+		writeChatCompletionsError(c, http.StatusBadGateway, "upstream_error", "Upstream request failed")
+		return nil, fmt.Errorf("upstream request failed: %s", safeErr)
+	}
+	defer func() { _ = resp.Body.Close() }()
+
+	// 7. Handle error response with failover
+	if resp.StatusCode >= 400 {
+		respBody, _ := io.ReadAll(io.LimitReader(resp.Body, 2<<20))
+		_ = resp.Body.Close()
+		resp.Body = io.NopCloser(bytes.NewReader(respBody))
+
+		upstreamMsg := strings.TrimSpace(extractUpstreamErrorMessage(respBody))
+		upstreamMsg = sanitizeUpstreamErrorMessage(upstreamMsg)
+		if s.shouldFailoverOpenAIUpstreamResponse(resp.StatusCode, upstreamMsg, respBody) {
+			upstreamDetail := ""
+			if s.cfg != nil && s.cfg.Gateway.LogUpstreamErrorBody {
+				maxBytes := s.cfg.Gateway.LogUpstreamErrorBodyMaxBytes
+				if maxBytes <= 0 {
+					maxBytes = 2048
+				}
+				upstreamDetail = truncateString(string(respBody), maxBytes)
+			}
+			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
+				Platform:           account.Platform,
+				AccountID:          account.ID,
+				AccountName:        account.Name,
+				UpstreamStatusCode: resp.StatusCode,
+				UpstreamRequestID:  resp.Header.Get("x-request-id"),
+				Kind:               "failover",
+				Message:            upstreamMsg,
+				Detail:             upstreamDetail,
+			})
+			if s.rateLimitService != nil {
+				s.rateLimitService.HandleUpstreamError(ctx, account, resp.StatusCode, resp.Header, respBody)
+			}
+			return nil, &UpstreamFailoverError{
+				StatusCode:             resp.StatusCode,
+				ResponseBody:           respBody,
+				RetryableOnSameAccount: account.IsPoolMode() && (isPoolModeRetryableStatus(resp.StatusCode) || isOpenAITransientProcessingError(resp.StatusCode, upstreamMsg, respBody)),
+			}
+		}
+		return s.handleChatCompletionsErrorResponse(resp, c, account)
+	}
+
+	// 8. Forward response
+	if clientStream {
+		return s.streamRawChatCompletions(c, resp, originalModel, billingModel, upstreamModel, reasoningEffort, serviceTier, startTime)
+	}
+	return s.bufferRawChatCompletions(c, resp, originalModel, billingModel, upstreamModel, reasoningEffort, serviceTier, startTime)
+}
+
+// streamRawChatCompletions 透传上游 CC SSE 流到客户端，并提取 usage（包括
+// 末尾 [DONE] 之前的 chunk 中的 usage 字段，按 OpenAI CC 协议）。
+//
+// usage 字段仅在客户端请求 stream_options.include_usage=true 时出现于上游响应中。
+// 本函数不检查客户端的请求 flag——上游会自行处理，我们仅在上游响应
+// chunk 中出现 usage 时提取。
+func (s *OpenAIGatewayService) streamRawChatCompletions(
+	c *gin.Context,
+	resp *http.Response,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	reasoningEffort *string,
+	serviceTier *string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	c.Writer.Header().Set("Cache-Control", "no-cache")
+	c.Writer.Header().Set("Connection", "keep-alive")
+	c.Writer.Header().Set("X-Accel-Buffering", "no")
+	c.Writer.WriteHeader(http.StatusOK)
+
+	scanner := bufio.NewScanner(resp.Body)
+	maxLineSize := defaultMaxLineSize
+	if s.cfg != nil && s.cfg.Gateway.MaxLineSize > 0 {
+		maxLineSize = s.cfg.Gateway.MaxLineSize
+	}
+	scanner.Buffer(make([]byte, 0, 64*1024), maxLineSize)
+
+	var usage OpenAIUsage
+	var firstTokenMs *int
+
+	for scanner.Scan() {
+		line := scanner.Text()
+		// Direct passthrough: write each line + blank line separator
+		if _, werr := c.Writer.WriteString(line + "\n"); werr != nil {
+			logger.L().Debug("openai chat_completions raw: client write failed",
+				zap.Error(werr),
+				zap.String("request_id", requestID),
+			)
+			break
+		}
+		if line == "" {
+			c.Writer.Flush()
+			continue
+		}
+		c.Writer.Flush()
+
+		// Track first token timing on first non-empty data line
+		if firstTokenMs == nil && strings.HasPrefix(line, "data: ") && line != "data: [DONE]" {
+			elapsed := int(time.Since(startTime).Milliseconds())
+			firstTokenMs = &elapsed
+		}
+
+		// Extract usage from any chunk that carries it (CC streams typically put
+		// usage in the final chunk before [DONE], but may also appear elsewhere).
+		if strings.HasPrefix(line, "data: ") && line != "data: [DONE]" {
+			payload := line[6:]
+			if u := extractCCStreamUsage(payload); u != nil {
+				usage = *u
+			}
+		}
+	}
+
+	if err := scanner.Err(); err != nil {
+		if !errors.Is(err, context.Canceled) && !errors.Is(err, context.DeadlineExceeded) {
+			logger.L().Warn("openai chat_completions raw: stream read error",
+				zap.Error(err),
+				zap.String("request_id", requestID),
+			)
+		}
+	}
+
+	return &OpenAIForwardResult{
+		RequestID:       requestID,
+		Usage:           usage,
+		Model:           originalModel,
+		BillingModel:    billingModel,
+		UpstreamModel:   upstreamModel,
+		ReasoningEffort: reasoningEffort,
+		ServiceTier:     serviceTier,
+		Stream:          true,
+		Duration:        time.Since(startTime),
+		FirstTokenMs:    firstTokenMs,
+	}, nil
+}
+
+// bufferRawChatCompletions 透传上游 CC 非流式 JSON 响应。
+func (s *OpenAIGatewayService) bufferRawChatCompletions(
+	c *gin.Context,
+	resp *http.Response,
+	originalModel string,
+	billingModel string,
+	upstreamModel string,
+	reasoningEffort *string,
+	serviceTier *string,
+	startTime time.Time,
+) (*OpenAIForwardResult, error) {
+	requestID := resp.Header.Get("x-request-id")
+
+	respBody, err := io.ReadAll(io.LimitReader(resp.Body, 32<<20))
+	if err != nil {
+		writeChatCompletionsError(c, http.StatusBadGateway, "api_error", "Failed to read upstream response")
+		return nil, fmt.Errorf("read upstream body: %w", err)
+	}
+
+	var ccResp apicompat.ChatCompletionsResponse
+	var usage OpenAIUsage
+	if err := json.Unmarshal(respBody, &ccResp); err == nil && ccResp.Usage != nil {
+		usage = OpenAIUsage{
+			InputTokens:  ccResp.Usage.PromptTokens,
+			OutputTokens: ccResp.Usage.CompletionTokens,
+		}
+		if ccResp.Usage.PromptTokensDetails != nil {
+			usage.CacheReadInputTokens = ccResp.Usage.PromptTokensDetails.CachedTokens
+		}
+	}
+
+	if s.responseHeaderFilter != nil {
+		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
+	}
+	if ct := resp.Header.Get("Content-Type"); ct != "" {
+		c.Writer.Header().Set("Content-Type", ct)
+	} else {
+		c.Writer.Header().Set("Content-Type", "application/json")
+	}
+	c.Writer.WriteHeader(http.StatusOK)
+	_, _ = c.Writer.Write(respBody)
+
+	return &OpenAIForwardResult{
+		RequestID:       requestID,
+		Usage:           usage,
+		Model:           originalModel,
+		BillingModel:    billingModel,
+		UpstreamModel:   upstreamModel,
+		ReasoningEffort: reasoningEffort,
+		ServiceTier:     serviceTier,
+		Stream:          false,
+		Duration:        time.Since(startTime),
+	}, nil
+}
+
+// extractCCStreamUsage 从单个 CC 流式 chunk 的 payload 中提取 usage 字段。
+// CC 协议中 usage 仅出现在末尾 chunk（且仅当客户端请求 stream_options.include_usage
+// 时），但上游可能在多个 chunk 中重复——总是用最新值。
+func extractCCStreamUsage(payload string) *OpenAIUsage {
+	usageResult := gjson.Get(payload, "usage")
+	if !usageResult.Exists() || !usageResult.IsObject() {
+		return nil
+	}
+	u := OpenAIUsage{
+		InputTokens:  int(gjson.Get(payload, "usage.prompt_tokens").Int()),
+		OutputTokens: int(gjson.Get(payload, "usage.completion_tokens").Int()),
+	}
+	if cached := gjson.Get(payload, "usage.prompt_tokens_details.cached_tokens"); cached.Exists() {
+		u.CacheReadInputTokens = int(cached.Int())
+	}
+	return &u
+}
+
+// buildOpenAIChatCompletionsURL 拼接上游 Chat Completions 端点 URL。
+//
+//   - base 已是 /chat/completions：原样返回
+//   - base 以 /v1 结尾：追加 /chat/completions
+//   - 其他情况：追加 /v1/chat/completions
+//
+// 与 buildOpenAIResponsesURL 是姐妹函数。
+func buildOpenAIChatCompletionsURL(base string) string {
+	normalized := strings.TrimRight(strings.TrimSpace(base), "/")
+	if strings.HasSuffix(normalized, "/chat/completions") {
+		return normalized
+	}
+	if strings.HasSuffix(normalized, "/v1") {
+		return normalized + "/chat/completions"
+	}
+	return normalized + "/v1/chat/completions"
+}

backend/internal/service/openai_gateway_chat_completions_raw_test.go

@@ -0,0 +1,67 @@
+//go:build unit
+
+package service
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+)
+
+func TestBuildOpenAIChatCompletionsURL(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		base string
+		want string
+	}{
+		// 已是 /chat/completions：原样返回
+		{"already chat/completions", "https://api.openai.com/v1/chat/completions", "https://api.openai.com/v1/chat/completions"},
+		// 以 /v1 结尾：追加 /chat/completions
+		{"bare /v1", "https://api.openai.com/v1", "https://api.openai.com/v1/chat/completions"},
+		// 其他情况：追加 /v1/chat/completions
+		{"bare domain", "https://api.openai.com", "https://api.openai.com/v1/chat/completions"},
+		{"domain with trailing slash", "https://api.openai.com/", "https://api.openai.com/v1/chat/completions"},
+		// 第三方上游常见形式
+		{"third-party bare domain", "https://api.deepseek.com", "https://api.deepseek.com/v1/chat/completions"},
+		{"third-party with path prefix", "https://api.gptgod.online/api", "https://api.gptgod.online/api/v1/chat/completions"},
+		// 带空白字符
+		{"whitespace trimmed", "  https://api.openai.com/v1  ", "https://api.openai.com/v1/chat/completions"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			got := buildOpenAIChatCompletionsURL(tt.base)
+			require.Equal(t, tt.want, got)
+		})
+	}
+}
+
+// TestBuildOpenAIResponsesURL_ProbeURL 锁定 probe/测试端点使用的 URL 构建逻辑，
+// 确保 buildOpenAIResponsesURL 对标准 OpenAI base_url 格式均拼出 `/v1/responses`。
+func TestBuildOpenAIResponsesURL_ProbeURL(t *testing.T) {
+	t.Parallel()
+
+	tests := []struct {
+		name string
+		base string
+		want string
+	}{
+		{"bare domain", "https://api.openai.com", "https://api.openai.com/v1/responses"},
+		{"domain trailing slash", "https://api.openai.com/", "https://api.openai.com/v1/responses"},
+		{"bare /v1", "https://api.openai.com/v1", "https://api.openai.com/v1/responses"},
+		{"already /responses", "https://api.openai.com/v1/responses", "https://api.openai.com/v1/responses"},
+		{"third-party bare domain", "https://api.deepseek.com", "https://api.deepseek.com/v1/responses"},
+		{"only domain, no scheme", "api.gptgod.online", "api.gptgod.online/v1/responses"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			got := buildOpenAIResponsesURL(tt.base)
+			require.Equal(t, tt.want, got)
+		})
+	}
+}