diff --git a/backend/internal/service/openai_gateway_chat_completions.go b/backend/internal/service/openai_gateway_chat_completions.go
index 646c886047..403752baf5 100644
--- a/backend/internal/service/openai_gateway_chat_completions.go
+++ b/backend/internal/service/openai_gateway_chat_completions.go
@@ -331,6 +331,11 @@ func (s *OpenAIGatewayService) forwardAsChatCompletions(
 
 	// 6. Build upstream request
 	upstreamCtx, releaseUpstreamCtx := detachUpstreamContext(ctx)
+	cancelUpstream := func() {}
+	if clientStream {
+		upstreamCtx, cancelUpstream = context.WithCancel(upstreamCtx)
+	}
+	defer cancelUpstream()
 	upstreamReq, err := s.buildUpstreamRequest(upstreamCtx, c, account, responsesBody, token, true, promptCacheKey, false)
 	releaseUpstreamCtx()
 	if err != nil {
@@ -355,7 +360,10 @@ func (s *OpenAIGatewayService) forwardAsChatCompletions(
 	if err != nil {
 		return nil, s.handleOpenAIUpstreamTransportError(ctx, c, account, err, false)
 	}
-	defer func() { _ = resp.Body.Close() }()
+	defer func() {
+		cancelUpstream()
+		_ = resp.Body.Close()
+	}()
 
 	// 8. Handle error response with failover
 	if resp.StatusCode >= 400 {
diff --git a/backend/internal/service/openai_gateway_chat_completions_cancellation_test.go b/backend/internal/service/openai_gateway_chat_completions_cancellation_test.go
new file mode 100644
index 0000000000..4115dc65b4
--- /dev/null
+++ b/backend/internal/service/openai_gateway_chat_completions_cancellation_test.go
@@ -0,0 +1,116 @@
+//go:build unit
+
+package service
+
+import (
+	"context"
+	"io"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/Wei-Shaw/sub2api/internal/pkg/tlsfingerprint"
+	"github.com/gin-gonic/gin"
+	"github.com/stretchr/testify/require"
+)
+
+type contextBoundBlockingReadCloser struct {
+	data       []byte
+	offset     int
+	ctx        context.Context
+	forceClose chan struct{}
+	closeOnce  sync.Once
+}
+
+func newContextBoundBlockingReadCloser(data []byte) *contextBoundBlockingReadCloser {
+	return &contextBoundBlockingReadCloser{
+		data:       data,
+		forceClose: make(chan struct{}),
+	}
+}
+
+func (r *contextBoundBlockingReadCloser) Read(p []byte) (int, error) {
+	if r.offset < len(r.data) {
+		n := copy(p, r.data[r.offset:])
+		r.offset += n
+		return n, nil
+	}
+	select {
+	case <-r.ctx.Done():
+		return 0, r.ctx.Err()
+	case <-r.forceClose:
+		return 0, io.EOF
+	}
+}
+
+func (r *contextBoundBlockingReadCloser) Close() error {
+	select {
+	case <-r.ctx.Done():
+	case <-r.forceClose:
+	}
+	return nil
+}
+
+func (r *contextBoundBlockingReadCloser) forceUnblock() {
+	r.closeOnce.Do(func() { close(r.forceClose) })
+}
+
+type contextBoundHTTPUpstream struct {
+	body *contextBoundBlockingReadCloser
+}
+
+func (u *contextBoundHTTPUpstream) Do(req *http.Request, _ string, _ int64, _ int) (*http.Response, error) {
+	u.body.ctx = req.Context()
+	return &http.Response{
+		StatusCode: http.StatusOK,
+		Header:     http.Header{"Content-Type": []string{"text/event-stream"}},
+		Body:       u.body,
+	}, nil
+}
+
+func (u *contextBoundHTTPUpstream) DoWithTLS(req *http.Request, proxyURL string, accountID int64, accountConcurrency int, _ *tlsfingerprint.Profile) (*http.Response, error) {
+	return u.Do(req, proxyURL, accountID, accountConcurrency)
+}
+
+func TestForwardAsChatCompletions_CancelsUpstreamBeforeClosingBody(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	rec := httptest.NewRecorder()
+	c, _ := gin.CreateTestContext(rec)
+	body := []byte(`{"model":"gpt-5.4","messages":[{"role":"user","content":"hello"}],"stream":true}`)
+	c.Request = httptest.NewRequest(http.MethodPost, "/v1/chat/completions", strings.NewReader(string(body)))
+	c.Request.Header.Set("Content-Type", "application/json")
+
+	upstreamBody := []byte("data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_1\",\"object\":\"response\",\"model\":\"gpt-5.4\",\"status\":\"completed\",\"output\":[{\"type\":\"message\",\"id\":\"msg_1\",\"role\":\"assistant\",\"status\":\"completed\",\"content\":[{\"type\":\"output_text\",\"text\":\"ok\"}]}],\"usage\":{\"input_tokens\":17,\"output_tokens\":8,\"total_tokens\":25}}}\n\n")
+	stream := newContextBoundBlockingReadCloser(upstreamBody)
+	t.Cleanup(stream.forceUnblock)
+
+	cfg := rawChatCompletionsTestConfig()
+	cfg.Gateway.StreamKeepaliveInterval = 1
+	svc := &OpenAIGatewayService{
+		cfg:          cfg,
+		httpUpstream: &contextBoundHTTPUpstream{body: stream},
+	}
+
+	type forwardResult struct {
+		result *OpenAIForwardResult
+		err    error
+	}
+	resultCh := make(chan forwardResult, 1)
+	go func() {
+		result, err := svc.ForwardAsChatCompletions(context.Background(), c, rawChatCompletionsTestAccount(), body, "", "gpt-5.1")
+		resultCh <- forwardResult{result: result, err: err}
+	}()
+
+	select {
+	case got := <-resultCh:
+		require.NoError(t, got.err)
+		require.NotNil(t, got.result)
+		require.Equal(t, 17, got.result.Usage.InputTokens)
+	case <-time.After(time.Second):
+		t.Fatal("ForwardAsChatCompletions did not cancel upstream before closing the body")
+	}
+}