Merge pull request #1248 from shekohex/feat/responses-compact

feat(openai): add responses/compact support

internal/api/server.go

@@ -326,6 +326,7 @@ func (s *Server) setupRoutes() {
 		v1.POST("/messages", claudeCodeHandlers.ClaudeMessages)
 		v1.POST("/messages/count_tokens", claudeCodeHandlers.ClaudeCountTokens)
 		v1.POST("/responses", openaiResponsesHandlers.Responses)
+		v1.POST("/responses/compact", openaiResponsesHandlers.Compact)
 	}
 
 	// Gemini compatible API routes

internal/runtime/executor/aistudio_executor.go

@@ -111,6 +111,9 @@ func (e *AIStudioExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.A
 
 // Execute performs a non-streaming request to the AI Studio API.
 func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
 	defer reporter.trackFailure(ctx, &err)
@@ -167,6 +170,9 @@ func (e *AIStudioExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth,
 
 // ExecuteStream performs a streaming request to the AI Studio API.
 func (e *AIStudioExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
 	defer reporter.trackFailure(ctx, &err)

internal/runtime/executor/antigravity_executor.go

@@ -109,6 +109,9 @@ func (e *AntigravityExecutor) HttpRequest(ctx context.Context, auth *cliproxyaut
 
 // Execute performs a non-streaming request to the Antigravity API.
 func (e *AntigravityExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 	isClaude := strings.Contains(strings.ToLower(baseModel), "claude")
 
@@ -641,6 +644,9 @@ func (e *AntigravityExecutor) convertStreamToNonStream(stream []byte) []byte {
 
 // ExecuteStream performs a streaming request to the Antigravity API.
 func (e *AntigravityExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	ctx = context.WithValue(ctx, "alt", "")

internal/runtime/executor/claude_executor.go

@@ -84,6 +84,9 @@ func (e *ClaudeExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Aut
 }
 
 func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, baseURL := claudeCreds(auth)
@@ -221,6 +224,9 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 }
 
 func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, baseURL := claudeCreds(auth)

internal/runtime/executor/codex_executor.go

@@ -73,6 +73,9 @@ func (e *CodexExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth
 }
 
 func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return e.executeCompact(ctx, auth, req, opts)
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, baseURL := codexCreds(auth)
@@ -117,7 +120,7 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	if err != nil {
 		return resp, err
 	}
-	applyCodexHeaders(httpReq, auth, apiKey)
+	applyCodexHeaders(httpReq, auth, apiKey, true)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -185,7 +188,96 @@ func (e *CodexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 	return resp, err
 }
 
+func (e *CodexExecutor) executeCompact(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	baseModel := thinking.ParseSuffix(req.Model).ModelName
+
+	apiKey, baseURL := codexCreds(auth)
+	if baseURL == "" {
+		baseURL = "https://chatgpt.com/backend-api/codex"
+	}
+
+	reporter := newUsageReporter(ctx, e.Identifier(), baseModel, auth)
+	defer reporter.trackFailure(ctx, &err)
+
+	from := opts.SourceFormat
+	to := sdktranslator.FromString("openai-response")
+	originalPayload := bytes.Clone(req.Payload)
+	if len(opts.OriginalRequest) > 0 {
+		originalPayload = bytes.Clone(opts.OriginalRequest)
+	}
+	originalTranslated := sdktranslator.TranslateRequest(from, to, baseModel, originalPayload, false)
+	body := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), false)
+
+	body, err = thinking.ApplyThinking(body, req.Model, from.String(), to.String(), e.Identifier())
+	if err != nil {
+		return resp, err
+	}
+
+	requestedModel := payloadRequestedModel(opts, req.Model)
+	body = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", body, originalTranslated, requestedModel)
+	body, _ = sjson.SetBytes(body, "model", baseModel)
+	body, _ = sjson.DeleteBytes(body, "stream")
+
+	url := strings.TrimSuffix(baseURL, "/") + "/responses/compact"
+	httpReq, err := e.cacheHelper(ctx, from, url, req, body)
+	if err != nil {
+		return resp, err
+	}
+	applyCodexHeaders(httpReq, auth, apiKey, false)
+	var authID, authLabel, authType, authValue string
+	if auth != nil {
+		authID = auth.ID
+		authLabel = auth.Label
+		authType, authValue = auth.AccountInfo()
+	}
+	recordAPIRequest(ctx, e.cfg, upstreamRequestLog{
+		URL:       url,
+		Method:    http.MethodPost,
+		Headers:   httpReq.Header.Clone(),
+		Body:      body,
+		Provider:  e.Identifier(),
+		AuthID:    authID,
+		AuthLabel: authLabel,
+		AuthType:  authType,
+		AuthValue: authValue,
+	})
+	httpClient := newProxyAwareHTTPClient(ctx, e.cfg, auth, 0)
+	httpResp, err := httpClient.Do(httpReq)
+	if err != nil {
+		recordAPIResponseError(ctx, e.cfg, err)
+		return resp, err
+	}
+	defer func() {
+		if errClose := httpResp.Body.Close(); errClose != nil {
+			log.Errorf("codex executor: close response body error: %v", errClose)
+		}
+	}()
+	recordAPIResponseMetadata(ctx, e.cfg, httpResp.StatusCode, httpResp.Header.Clone())
+	if httpResp.StatusCode < 200 || httpResp.StatusCode >= 300 {
+		b, _ := io.ReadAll(httpResp.Body)
+		appendAPIResponseChunk(ctx, e.cfg, b)
+		logWithRequestID(ctx).Debugf("request error, error status: %d, error message: %s", httpResp.StatusCode, summarizeErrorBody(httpResp.Header.Get("Content-Type"), b))
+		err = statusErr{code: httpResp.StatusCode, msg: string(b)}
+		return resp, err
+	}
+	data, err := io.ReadAll(httpResp.Body)
+	if err != nil {
+		recordAPIResponseError(ctx, e.cfg, err)
+		return resp, err
+	}
+	appendAPIResponseChunk(ctx, e.cfg, data)
+	reporter.publish(ctx, parseOpenAIUsage(data))
+	reporter.ensurePublished(ctx)
+	var param any
+	out := sdktranslator.TranslateNonStream(ctx, to, from, req.Model, bytes.Clone(originalPayload), body, data, &param)
+	resp = cliproxyexecutor.Response{Payload: []byte(out)}
+	return resp, nil
+}
+
 func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusBadRequest, msg: "streaming not supported for /responses/compact"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, baseURL := codexCreds(auth)
@@ -229,7 +321,7 @@ func (e *CodexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Au
 	if err != nil {
 		return nil, err
 	}
-	applyCodexHeaders(httpReq, auth, apiKey)
+	applyCodexHeaders(httpReq, auth, apiKey, true)
 	var authID, authLabel, authType, authValue string
 	if auth != nil {
 		authID = auth.ID
@@ -530,17 +622,21 @@ func (e *CodexExecutor) cacheHelper(ctx context.Context, from sdktranslator.Form
 		}
 	}
 
-	rawJSON, _ = sjson.SetBytes(rawJSON, "prompt_cache_key", cache.ID)
+	if cache.ID != "" {
+		rawJSON, _ = sjson.SetBytes(rawJSON, "prompt_cache_key", cache.ID)
+	}
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(rawJSON))
 	if err != nil {
 		return nil, err
 	}
-	httpReq.Header.Set("Conversation_id", cache.ID)
-	httpReq.Header.Set("Session_id", cache.ID)
+	if cache.ID != "" {
+		httpReq.Header.Set("Conversation_id", cache.ID)
+		httpReq.Header.Set("Session_id", cache.ID)
+	}
 	return httpReq, nil
 }
 
-func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string) {
+func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string, stream bool) {
 	r.Header.Set("Content-Type", "application/json")
 	r.Header.Set("Authorization", "Bearer "+token)
 
@@ -554,7 +650,11 @@ func applyCodexHeaders(r *http.Request, auth *cliproxyauth.Auth, token string) {
 	misc.EnsureHeader(r.Header, ginHeaders, "Session_id", uuid.NewString())
 	misc.EnsureHeader(r.Header, ginHeaders, "User-Agent", "codex_cli_rs/0.50.0 (Mac OS 26.0.1; arm64) Apple_Terminal/464")
 
-	r.Header.Set("Accept", "text/event-stream")
+	if stream {
+		r.Header.Set("Accept", "text/event-stream")
+	} else {
+		r.Header.Set("Accept", "application/json")
+	}
 	r.Header.Set("Connection", "Keep-Alive")
 
 	isAPIKey := false

internal/runtime/executor/gemini_cli_executor.go

@@ -103,6 +103,9 @@ func (e *GeminiCLIExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.
 
 // Execute performs a non-streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)
@@ -253,6 +256,9 @@ func (e *GeminiCLIExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth
 
 // ExecuteStream performs a streaming request to the Gemini CLI API.
 func (e *GeminiCLIExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	tokenSource, baseTokenData, err := prepareGeminiCLITokenSource(ctx, e.cfg, auth)

internal/runtime/executor/gemini_executor.go

@@ -103,6 +103,9 @@ func (e *GeminiExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Aut
 //   - cliproxyexecutor.Response: The response from the API
 //   - error: An error if the request fails
 func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, bearer := geminiCreds(auth)
@@ -207,6 +210,9 @@ func (e *GeminiExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 
 // ExecuteStream performs a streaming request to the Gemini API.
 func (e *GeminiExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, bearer := geminiCreds(auth)

internal/runtime/executor/gemini_vertex_executor.go

@@ -233,6 +233,9 @@ func (e *GeminiVertexExecutor) HttpRequest(ctx context.Context, auth *cliproxyau
 
 // Execute performs a non-streaming request to the Vertex AI API.
 func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	// Try API key authentication first
 	apiKey, baseURL := vertexAPICreds(auth)
 
@@ -251,6 +254,9 @@ func (e *GeminiVertexExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 
 // ExecuteStream performs a streaming request to the Vertex AI API.
 func (e *GeminiVertexExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	// Try API key authentication first
 	apiKey, baseURL := vertexAPICreds(auth)
 

internal/runtime/executor/iflow_executor.go

@@ -68,6 +68,9 @@ func (e *IFlowExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth
 
 // Execute performs a non-streaming chat completion request.
 func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, baseURL := iflowCreds(auth)
@@ -167,6 +170,9 @@ func (e *IFlowExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, re
 
 // ExecuteStream performs a streaming chat completion request.
 func (e *IFlowExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	apiKey, baseURL := iflowCreds(auth)

internal/runtime/executor/openai_compat_executor.go

@@ -81,9 +81,13 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 		return
 	}
 
-	// Translate inbound request to OpenAI format
 	from := opts.SourceFormat
 	to := sdktranslator.FromString("openai")
+	endpoint := "/chat/completions"
+	if opts.Alt == "responses/compact" {
+		to = sdktranslator.FromString("openai-response")
+		endpoint = "/responses/compact"
+	}
 	originalPayload := bytes.Clone(req.Payload)
 	if len(opts.OriginalRequest) > 0 {
 		originalPayload = bytes.Clone(opts.OriginalRequest)
@@ -92,13 +96,18 @@ func (e *OpenAICompatExecutor) Execute(ctx context.Context, auth *cliproxyauth.A
 	translated := sdktranslator.TranslateRequest(from, to, baseModel, bytes.Clone(req.Payload), opts.Stream)
 	requestedModel := payloadRequestedModel(opts, req.Model)
 	translated = applyPayloadConfigWithRoot(e.cfg, baseModel, to.String(), "", translated, originalTranslated, requestedModel)
+	if opts.Alt == "responses/compact" {
+		if updated, errDelete := sjson.DeleteBytes(translated, "stream"); errDelete == nil {
+			translated = updated
+		}
+	}
 
 	translated, err = thinking.ApplyThinking(translated, req.Model, from.String(), to.String(), e.Identifier())
 	if err != nil {
 		return resp, err
 	}
 
-	url := strings.TrimSuffix(baseURL, "/") + "/chat/completions"
+	url := strings.TrimSuffix(baseURL, "/") + endpoint
 	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(translated))
 	if err != nil {
 		return resp, err

internal/runtime/executor/openai_compat_executor_compact_test.go

@@ -0,0 +1,58 @@
+package executor
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
+	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
+	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
+	sdktranslator "github.com/router-for-me/CLIProxyAPI/v6/sdk/translator"
+	"github.com/tidwall/gjson"
+)
+
+func TestOpenAICompatExecutorCompactPassthrough(t *testing.T) {
+	var gotPath string
+	var gotBody []byte
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotPath = r.URL.Path
+		body, _ := io.ReadAll(r.Body)
+		gotBody = body
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"id":"resp_1","object":"response.compaction","usage":{"input_tokens":1,"output_tokens":2,"total_tokens":3}}`))
+	}))
+	defer server.Close()
+
+	executor := NewOpenAICompatExecutor("openai-compatibility", &config.Config{})
+	auth := &cliproxyauth.Auth{Attributes: map[string]string{
+		"base_url": server.URL + "/v1",
+		"api_key":  "test",
+	}}
+	payload := []byte(`{"model":"gpt-5.1-codex-max","input":[{"role":"user","content":"hi"}]}`)
+	resp, err := executor.Execute(context.Background(), auth, cliproxyexecutor.Request{
+		Model:   "gpt-5.1-codex-max",
+		Payload: payload,
+	}, cliproxyexecutor.Options{
+		SourceFormat: sdktranslator.FromString("openai-response"),
+		Alt:          "responses/compact",
+		Stream:       false,
+	})
+	if err != nil {
+		t.Fatalf("Execute error: %v", err)
+	}
+	if gotPath != "/v1/responses/compact" {
+		t.Fatalf("path = %q, want %q", gotPath, "/v1/responses/compact")
+	}
+	if !gjson.GetBytes(gotBody, "input").Exists() {
+		t.Fatalf("expected input in body")
+	}
+	if gjson.GetBytes(gotBody, "messages").Exists() {
+		t.Fatalf("unexpected messages in body")
+	}
+	if string(resp.Payload) != `{"id":"resp_1","object":"response.compaction","usage":{"input_tokens":1,"output_tokens":2,"total_tokens":3}}` {
+		t.Fatalf("payload = %s", string(resp.Payload))
+	}
+}

internal/runtime/executor/qwen_executor.go

@@ -66,6 +66,9 @@ func (e *QwenExecutor) HttpRequest(ctx context.Context, auth *cliproxyauth.Auth,
 }
 
 func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (resp cliproxyexecutor.Response, err error) {
+	if opts.Alt == "responses/compact" {
+		return resp, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	token, baseURL := qwenCreds(auth)
@@ -153,6 +156,9 @@ func (e *QwenExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, req
 }
 
 func (e *QwenExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.Auth, req cliproxyexecutor.Request, opts cliproxyexecutor.Options) (stream <-chan cliproxyexecutor.StreamChunk, err error) {
+	if opts.Alt == "responses/compact" {
+		return nil, statusErr{code: http.StatusNotImplemented, msg: "/responses/compact not supported"}
+	}
 	baseModel := thinking.ParseSuffix(req.Model).ModelName
 
 	token, baseURL := qwenCreds(auth)

internal/runtime/executor/usage_helpers.go

@@ -199,15 +199,31 @@ func parseOpenAIUsage(data []byte) usage.Detail {
 	if !usageNode.Exists() {
 		return usage.Detail{}
 	}
+	inputNode := usageNode.Get("prompt_tokens")
+	if !inputNode.Exists() {
+		inputNode = usageNode.Get("input_tokens")
+	}
+	outputNode := usageNode.Get("completion_tokens")
+	if !outputNode.Exists() {
+		outputNode = usageNode.Get("output_tokens")
+	}
 	detail := usage.Detail{
-		InputTokens:  usageNode.Get("prompt_tokens").Int(),
-		OutputTokens: usageNode.Get("completion_tokens").Int(),
+		InputTokens:  inputNode.Int(),
+		OutputTokens: outputNode.Int(),
 		TotalTokens:  usageNode.Get("total_tokens").Int(),
 	}
-	if cached := usageNode.Get("prompt_tokens_details.cached_tokens"); cached.Exists() {
+	cached := usageNode.Get("prompt_tokens_details.cached_tokens")
+	if !cached.Exists() {
+		cached = usageNode.Get("input_tokens_details.cached_tokens")
+	}
+	if cached.Exists() {
 		detail.CachedTokens = cached.Int()
 	}
-	if reasoning := usageNode.Get("completion_tokens_details.reasoning_tokens"); reasoning.Exists() {
+	reasoning := usageNode.Get("completion_tokens_details.reasoning_tokens")
+	if !reasoning.Exists() {
+		reasoning = usageNode.Get("output_tokens_details.reasoning_tokens")
+	}
+	if reasoning.Exists() {
 		detail.ReasoningTokens = reasoning.Int()
 	}
 	return detail

internal/runtime/executor/usage_helpers_test.go

@@ -0,0 +1,43 @@
+package executor
+
+import "testing"
+
+func TestParseOpenAIUsageChatCompletions(t *testing.T) {
+	data := []byte(`{"usage":{"prompt_tokens":1,"completion_tokens":2,"total_tokens":3,"prompt_tokens_details":{"cached_tokens":4},"completion_tokens_details":{"reasoning_tokens":5}}}`)
+	detail := parseOpenAIUsage(data)
+	if detail.InputTokens != 1 {
+		t.Fatalf("input tokens = %d, want %d", detail.InputTokens, 1)
+	}
+	if detail.OutputTokens != 2 {
+		t.Fatalf("output tokens = %d, want %d", detail.OutputTokens, 2)
+	}
+	if detail.TotalTokens != 3 {
+		t.Fatalf("total tokens = %d, want %d", detail.TotalTokens, 3)
+	}
+	if detail.CachedTokens != 4 {
+		t.Fatalf("cached tokens = %d, want %d", detail.CachedTokens, 4)
+	}
+	if detail.ReasoningTokens != 5 {
+		t.Fatalf("reasoning tokens = %d, want %d", detail.ReasoningTokens, 5)
+	}
+}
+
+func TestParseOpenAIUsageResponses(t *testing.T) {
+	data := []byte(`{"usage":{"input_tokens":10,"output_tokens":20,"total_tokens":30,"input_tokens_details":{"cached_tokens":7},"output_tokens_details":{"reasoning_tokens":9}}}`)
+	detail := parseOpenAIUsage(data)
+	if detail.InputTokens != 10 {
+		t.Fatalf("input tokens = %d, want %d", detail.InputTokens, 10)
+	}
+	if detail.OutputTokens != 20 {
+		t.Fatalf("output tokens = %d, want %d", detail.OutputTokens, 20)
+	}
+	if detail.TotalTokens != 30 {
+		t.Fatalf("total tokens = %d, want %d", detail.TotalTokens, 30)
+	}
+	if detail.CachedTokens != 7 {
+		t.Fatalf("cached tokens = %d, want %d", detail.CachedTokens, 7)
+	}
+	if detail.ReasoningTokens != 9 {
+		t.Fatalf("reasoning tokens = %d, want %d", detail.ReasoningTokens, 9)
+	}
+}