diff --git a/backend/internal/config/config.go b/backend/internal/config/config.go
index c25d13a8b4..18baa34881 100644
--- a/backend/internal/config/config.go
+++ b/backend/internal/config/config.go
@@ -721,6 +721,9 @@ type GatewayConfig struct {
 	// OpenAIPassthroughAllowTimeoutHeaders: OpenAI 透传模式是否放行客户端超时头
 	// 关闭（默认）可避免 x-stainless-timeout 等头导致上游提前断流。
 	OpenAIPassthroughAllowTimeoutHeaders bool `mapstructure:"openai_passthrough_allow_timeout_headers"`
+	// OpenAICompactModel: /responses/compact 上游使用的模型。
+	// compact 端点支持模型滞后于普通 /responses 时，可用该配置降级规避上游错误。
+	OpenAICompactModel string `mapstructure:"openai_compact_model"`
 	// OpenAIWS: OpenAI Responses WebSocket 配置（默认开启，可按需回滚到 HTTP）
 	OpenAIWS GatewayOpenAIWSConfig `mapstructure:"openai_ws"`
 	// OpenAIScheduler: OpenAI 高级调度器粘性逃逸配置
@@ -867,7 +870,7 @@ func (c *UserMessageQueueConfig) GetEffectiveMode() string {
 type GatewayOpenAIWSConfig struct {
 	// ModeRouterV2Enabled: 新版 WS mode 路由开关（默认 false；关闭时保持 legacy 行为）
 	ModeRouterV2Enabled bool `mapstructure:"mode_router_v2_enabled"`
-	// IngressModeDefault: ingress 默认模式（off/ctx_pool/passthrough）
+	// IngressModeDefault: ingress 默认模式（off/ctx_pool/passthrough/http_bridge）
 	IngressModeDefault string `mapstructure:"ingress_mode_default"`
 	// Enabled: 全局总开关（默认 true）
 	Enabled bool `mapstructure:"enabled"`
@@ -1836,6 +1839,7 @@ func setDefaults() {
 	viper.SetDefault("gateway.force_codex_cli", false)
 	viper.SetDefault("gateway.codex_image_generation_bridge_enabled", false)
 	viper.SetDefault("gateway.openai_passthrough_allow_timeout_headers", false)
+	viper.SetDefault("gateway.openai_compact_model", "gpt-5.4")
 	// OpenAI Responses WebSocket（默认开启；可通过 force_http 紧急回滚）
 	viper.SetDefault("gateway.openai_ws.enabled", true)
 	viper.SetDefault("gateway.openai_ws.mode_router_v2_enabled", false)
@@ -2624,11 +2628,11 @@ func (c *Config) Validate() error {
 	}
 	if mode := strings.ToLower(strings.TrimSpace(c.Gateway.OpenAIWS.IngressModeDefault)); mode != "" {
 		switch mode {
-		case "off", "ctx_pool", "passthrough":
+		case "off", "ctx_pool", "passthrough", "http_bridge":
 		case "shared", "dedicated":
-			slog.Warn("gateway.openai_ws.ingress_mode_default is deprecated, treating as ctx_pool; please update to off|ctx_pool|passthrough", "value", mode)
+			slog.Warn("gateway.openai_ws.ingress_mode_default is deprecated, treating as ctx_pool; please update to off|ctx_pool|passthrough|http_bridge", "value", mode)
 		default:
-			return fmt.Errorf("gateway.openai_ws.ingress_mode_default must be one of off|ctx_pool|passthrough")
+			return fmt.Errorf("gateway.openai_ws.ingress_mode_default must be one of off|ctx_pool|passthrough|http_bridge")
 		}
 	}
 	if mode := strings.ToLower(strings.TrimSpace(c.Gateway.OpenAIWS.StoreDisabledConnMode)); mode != "" {
diff --git a/backend/internal/config/config_test.go b/backend/internal/config/config_test.go
index be31976156..bf7a327563 100644
--- a/backend/internal/config/config_test.go
+++ b/backend/internal/config/config_test.go
@@ -184,6 +184,23 @@ func TestLoadDefaultOpenAIWSConfig(t *testing.T) {
 	}
 }
 
+func TestLoadDefaultOpenAICompactModel(t *testing.T) {
+	resetViperWithJWTSecret(t)
+
+	cfg, err := Load()
+	require.NoError(t, err)
+	require.Equal(t, "gpt-5.4", cfg.Gateway.OpenAICompactModel)
+}
+
+func TestLoadOpenAICompactModelFromEnv(t *testing.T) {
+	resetViperWithJWTSecret(t)
+	t.Setenv("GATEWAY_OPENAI_COMPACT_MODEL", "gpt-5.3-codex")
+
+	cfg, err := Load()
+	require.NoError(t, err)
+	require.Equal(t, "gpt-5.3-codex", cfg.Gateway.OpenAICompactModel)
+}
+
 func TestLoadDefaultOpenAIHTTP2Enabled(t *testing.T) {
 	resetViperWithJWTSecret(t)
 
@@ -1678,7 +1695,7 @@ func TestValidateConfig_OpenAIWSRules(t *testing.T) {
 			wantErr: "gateway.openai_ws.store_disabled_conn_mode",
 		},
 		{
-			name:    "ingress_mode_default 必须为 off|ctx_pool|passthrough",
+			name:    "ingress_mode_default 必须为 off|ctx_pool|passthrough|http_bridge",
 			mutate:  func(c *Config) { c.Gateway.OpenAIWS.IngressModeDefault = "invalid" },
 			wantErr: "gateway.openai_ws.ingress_mode_default",
 		},
diff --git a/backend/internal/handler/openai_gateway_handler.go b/backend/internal/handler/openai_gateway_handler.go
index ac346fea3f..8eec667fa6 100644
--- a/backend/internal/handler/openai_gateway_handler.go
+++ b/backend/internal/handler/openai_gateway_handler.go
@@ -1349,7 +1349,7 @@ func (h *OpenAIGatewayHandler) ResponsesWebSocket(c *gin.Context) {
 
 	subscription, _ := middleware2.GetSubscriptionFromContext(c)
 	requestPlatform := openAICompatibleRequestPlatform(apiKey)
-	requiredTransport := service.OpenAIUpstreamTransportResponsesWebsocketV2
+	requiredTransport := service.OpenAIUpstreamTransportResponsesWebsocketV2Ingress
 	if requestPlatform == service.PlatformGrok {
 		requiredTransport = service.OpenAIUpstreamTransportHTTPSSE
 	}
diff --git a/backend/internal/service/account.go b/backend/internal/service/account.go
index 7f68da46c6..7d5e70a15b 100644
--- a/backend/internal/service/account.go
+++ b/backend/internal/service/account.go
@@ -1474,6 +1474,7 @@ const (
 	OpenAIWSIngressModeDedicated   = "dedicated"
 	OpenAIWSIngressModeCtxPool     = "ctx_pool"
 	OpenAIWSIngressModePassthrough = "passthrough"
+	OpenAIWSIngressModeHTTPBridge  = "http_bridge"
 )
 
 func normalizeOpenAIWSIngressMode(mode string) string {
@@ -1484,6 +1485,8 @@ func normalizeOpenAIWSIngressMode(mode string) string {
 		return OpenAIWSIngressModeCtxPool
 	case OpenAIWSIngressModePassthrough:
 		return OpenAIWSIngressModePassthrough
+	case OpenAIWSIngressModeHTTPBridge:
+		return OpenAIWSIngressModeHTTPBridge
 	case OpenAIWSIngressModeShared:
 		return OpenAIWSIngressModeShared
 	case OpenAIWSIngressModeDedicated:
diff --git a/backend/internal/service/account_openai_passthrough_test.go b/backend/internal/service/account_openai_passthrough_test.go
index 50c2b7cb86..2605fdaeb3 100644
--- a/backend/internal/service/account_openai_passthrough_test.go
+++ b/backend/internal/service/account_openai_passthrough_test.go
@@ -229,6 +229,17 @@ func TestAccount_ResolveOpenAIResponsesWebSocketV2Mode(t *testing.T) {
 		require.Equal(t, OpenAIWSIngressModePassthrough, account.ResolveOpenAIResponsesWebSocketV2Mode(OpenAIWSIngressModeCtxPool))
 	})
 
+	t.Run("oauth mode supports http_bridge", func(t *testing.T) {
+		account := &Account{
+			Platform: PlatformOpenAI,
+			Type:     AccountTypeOAuth,
+			Extra: map[string]any{
+				"openai_oauth_responses_websockets_v2_mode": OpenAIWSIngressModeHTTPBridge,
+			},
+		}
+		require.Equal(t, OpenAIWSIngressModeHTTPBridge, account.ResolveOpenAIResponsesWebSocketV2Mode(OpenAIWSIngressModeCtxPool))
+	})
+
 	t.Run("legacy enabled maps to ctx_pool", func(t *testing.T) {
 		account := &Account{
 			Platform: PlatformOpenAI,
diff --git a/backend/internal/service/openai_account_scheduler.go b/backend/internal/service/openai_account_scheduler.go
index 42aa494b1f..ab298d4521 100644
--- a/backend/internal/service/openai_account_scheduler.go
+++ b/backend/internal/service/openai_account_scheduler.go
@@ -1420,6 +1420,18 @@ func (s *OpenAIGatewayService) isOpenAIAccountTransportCompatible(account *Accou
 	if s == nil || account == nil {
 		return false
 	}
+	if requiredTransport == OpenAIUpstreamTransportResponsesWebsocketV2Ingress {
+		if s.cfg == nil || !s.cfg.Gateway.OpenAIWS.ModeRouterV2Enabled {
+			return s.getOpenAIWSProtocolResolver().Resolve(account).Transport == OpenAIUpstreamTransportResponsesWebsocketV2
+		}
+		mode := account.ResolveOpenAIResponsesWebSocketV2Mode(s.cfg.Gateway.OpenAIWS.IngressModeDefault)
+		switch mode {
+		case OpenAIWSIngressModeCtxPool, OpenAIWSIngressModePassthrough, OpenAIWSIngressModeHTTPBridge, OpenAIWSIngressModeShared, OpenAIWSIngressModeDedicated:
+			return true
+		default:
+			return false
+		}
+	}
 	return s.getOpenAIWSProtocolResolver().Resolve(account).Transport == requiredTransport
 }
 
diff --git a/backend/internal/service/openai_account_scheduler_test.go b/backend/internal/service/openai_account_scheduler_test.go
index 6d8e38d0bc..0255dbbde3 100644
--- a/backend/internal/service/openai_account_scheduler_test.go
+++ b/backend/internal/service/openai_account_scheduler_test.go
@@ -2391,6 +2391,7 @@ func TestDefaultOpenAIAccountScheduler_IsAccountTransportCompatible_Branches(t *
 	require.True(t, scheduler.isAccountTransportCompatible(nil, OpenAIUpstreamTransportAny))
 	require.True(t, scheduler.isAccountTransportCompatible(nil, OpenAIUpstreamTransportHTTPSSE))
 	require.False(t, scheduler.isAccountTransportCompatible(nil, OpenAIUpstreamTransportResponsesWebsocketV2))
+	require.False(t, scheduler.isAccountTransportCompatible(nil, OpenAIUpstreamTransportResponsesWebsocketV2Ingress))
 
 	cfg := newSchedulerTestOpenAIWSV2Config()
 	scheduler.service = &OpenAIGatewayService{cfg: cfg}
@@ -2406,6 +2407,15 @@ func TestDefaultOpenAIAccountScheduler_IsAccountTransportCompatible_Branches(t *
 		},
 	}
 	require.True(t, scheduler.isAccountTransportCompatible(account, OpenAIUpstreamTransportResponsesWebsocketV2))
+	require.True(t, scheduler.isAccountTransportCompatible(account, OpenAIUpstreamTransportResponsesWebsocketV2Ingress))
+
+	cfg.Gateway.OpenAIWS.ModeRouterV2Enabled = true
+	account.Extra["openai_apikey_responses_websockets_v2_mode"] = OpenAIWSIngressModeHTTPBridge
+	require.False(t, scheduler.isAccountTransportCompatible(account, OpenAIUpstreamTransportResponsesWebsocketV2))
+	require.True(t, scheduler.isAccountTransportCompatible(account, OpenAIUpstreamTransportResponsesWebsocketV2Ingress))
+
+	account.Extra["openai_apikey_responses_websockets_v2_mode"] = OpenAIWSIngressModeOff
+	require.False(t, scheduler.isAccountTransportCompatible(account, OpenAIUpstreamTransportResponsesWebsocketV2Ingress))
 }
 
 func int64PtrForTest(v int64) *int64 {
diff --git a/backend/internal/service/openai_oauth_passthrough_test.go b/backend/internal/service/openai_oauth_passthrough_test.go
index 8d326b7637..3fbdbf8534 100644
--- a/backend/internal/service/openai_oauth_passthrough_test.go
+++ b/backend/internal/service/openai_oauth_passthrough_test.go
@@ -3,6 +3,7 @@ package service
 import (
 	"bytes"
 	"context"
+	"errors"
 	"fmt"
 	"io"
 	"net/http"
@@ -35,6 +36,21 @@ type httpUpstreamRecorder struct {
 	err       error
 }
 
+type passthroughErrReadCloser struct {
+	err error
+}
+
+func (r passthroughErrReadCloser) Read(_ []byte) (int, error) {
+	if r.err != nil {
+		return 0, r.err
+	}
+	return 0, io.ErrUnexpectedEOF
+}
+
+func (r passthroughErrReadCloser) Close() error {
+	return nil
+}
+
 func (u *httpUpstreamRecorder) Do(req *http.Request, proxyURL string, accountID int64, accountConcurrency int) (*http.Response, error) {
 	u.lastReq = req
 	u.lastProxyURL = proxyURL
@@ -799,7 +815,7 @@ func TestOpenAIGatewayService_OAuthPassthrough_UpstreamErrorIncludesPassthroughF
 	require.Equal(t, "http_error", arr[len(arr)-1].Kind)
 }
 
-func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *testing.T) {
+func TestOpenAIGatewayService_OpenAIPassthrough_RetryableStatusesTriggerFailover(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 	originalBody := []byte(`{"model":"gpt-5.2","stream":false,"instructions":"local-test-instructions","input":[{"type":"text","text":"hi"}]}`)
 
@@ -825,11 +841,12 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 	}
 
 	testCases := []struct {
-		name        string
-		accountType string
-		statusCode  int
-		body        string
-		assertRepo  func(t *testing.T, repo *openAIPassthroughFailoverRepo, start time.Time)
+		name           string
+		accountType    string
+		statusCode     int
+		body           string
+		expectFailover bool
+		assertRepo     func(t *testing.T, repo *openAIPassthroughFailoverRepo, start time.Time)
 	}{
 		{
 			name:        "oauth_429_rate_limit",
@@ -839,6 +856,7 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 				resetAt := time.Now().Add(7 * 24 * time.Hour).Unix()
 				return fmt.Sprintf(`{"error":{"message":"The usage limit has been reached","type":"usage_limit_reached","resets_at":%d}}`, resetAt)
 			}(),
+			expectFailover: true,
 			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, _ time.Time) {
 				require.Len(t, repo.rateLimitCalls, 1)
 				require.Empty(t, repo.overloadCalls)
@@ -846,16 +864,50 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 			},
 		},
 		{
-			name:        "oauth_529_overload",
-			accountType: AccountTypeOAuth,
-			statusCode:  529,
-			body:        `{"error":{"message":"server overloaded","type":"server_error"}}`,
+			name:           "oauth_529_overload",
+			accountType:    AccountTypeOAuth,
+			statusCode:     529,
+			body:           `{"error":{"message":"server overloaded","type":"server_error"}}`,
+			expectFailover: true,
 			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, start time.Time) {
 				require.Empty(t, repo.rateLimitCalls)
 				require.Len(t, repo.overloadCalls, 1)
 				require.WithinDuration(t, start.Add(10*time.Minute), repo.overloadCalls[0], 5*time.Second)
 			},
 		},
+		{
+			name:           "oauth_502_bad_gateway",
+			accountType:    AccountTypeOAuth,
+			statusCode:     http.StatusBadGateway,
+			body:           `{"error":{"message":"bad gateway","type":"server_error"}}`,
+			expectFailover: false,
+			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, _ time.Time) {
+				require.Empty(t, repo.rateLimitCalls)
+				require.Empty(t, repo.overloadCalls)
+			},
+		},
+		{
+			name:           "oauth_503_unavailable",
+			accountType:    AccountTypeOAuth,
+			statusCode:     http.StatusServiceUnavailable,
+			body:           `{"error":{"message":"service unavailable","type":"server_error"}}`,
+			expectFailover: false,
+			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, _ time.Time) {
+				require.Empty(t, repo.rateLimitCalls)
+				require.Empty(t, repo.overloadCalls)
+			},
+		},
+		{
+			name:           "oauth_504_gateway_timeout",
+			accountType:    AccountTypeOAuth,
+			statusCode:     http.StatusGatewayTimeout,
+			body:           `{"error":{"message":"gateway timeout","type":"server_error"}}`,
+			expectFailover: false,
+			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, _ time.Time) {
+				require.Empty(t, repo.rateLimitCalls)
+				require.Empty(t, repo.overloadCalls)
+			},
+		},
 		{
 			name:        "apikey_429_rate_limit",
 			accountType: AccountTypeAPIKey,
@@ -864,6 +916,7 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 				resetAt := time.Now().Add(7 * 24 * time.Hour).Unix()
 				return fmt.Sprintf(`{"error":{"message":"The usage limit has been reached","type":"usage_limit_reached","resets_at":%d}}`, resetAt)
 			}(),
+			expectFailover: true,
 			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, _ time.Time) {
 				require.Len(t, repo.rateLimitCalls, 1)
 				require.Empty(t, repo.overloadCalls)
@@ -871,10 +924,11 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 			},
 		},
 		{
-			name:        "apikey_529_overload",
-			accountType: AccountTypeAPIKey,
-			statusCode:  529,
-			body:        `{"error":{"message":"server overloaded","type":"server_error"}}`,
+			name:           "apikey_529_overload",
+			accountType:    AccountTypeAPIKey,
+			statusCode:     529,
+			body:           `{"error":{"message":"server overloaded","type":"server_error"}}`,
+			expectFailover: true,
 			assertRepo: func(t *testing.T, repo *openAIPassthroughFailoverRepo, start time.Time) {
 				require.Empty(t, repo.rateLimitCalls)
 				require.Len(t, repo.overloadCalls, 1)
@@ -919,9 +973,15 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 			require.Error(t, err)
 
 			var failoverErr *UpstreamFailoverError
-			require.ErrorAs(t, err, &failoverErr)
-			require.Equal(t, tc.statusCode, failoverErr.StatusCode)
-			require.False(t, c.Writer.Written(), "429/529 passthrough 应返回 failover 错误给上层换号，而不是直接向客户端写响应")
+			if tc.expectFailover {
+				require.ErrorAs(t, err, &failoverErr)
+				require.Equal(t, tc.statusCode, failoverErr.StatusCode)
+				require.False(t, c.Writer.Written(), "retryable passthrough 错误应返回 failover 错误给上层换号，而不是直接向客户端写响应")
+			} else {
+				require.False(t, errors.As(err, &failoverErr))
+				require.True(t, c.Writer.Written(), "非 failover 的 passthrough http 错误应直接写回客户端")
+				require.Equal(t, tc.statusCode, rec.Code)
+			}
 
 			v, ok := c.Get(OpsUpstreamErrorsKey)
 			require.True(t, ok)
@@ -929,7 +989,11 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 			require.True(t, ok)
 			require.NotEmpty(t, arr)
 			require.True(t, arr[len(arr)-1].Passthrough)
-			require.Equal(t, "failover", arr[len(arr)-1].Kind)
+			if tc.expectFailover {
+				require.Equal(t, "failover", arr[len(arr)-1].Kind)
+			} else {
+				require.Equal(t, "http_error", arr[len(arr)-1].Kind)
+			}
 			require.Equal(t, tc.statusCode, arr[len(arr)-1].UpstreamStatusCode)
 
 			tc.assertRepo(t, repo, start)
@@ -937,6 +1001,73 @@ func TestOpenAIGatewayService_OpenAIPassthrough_429And529TriggerFailover(t *test
 	}
 }
 
+func TestOpenAIGatewayService_OpenAIPassthrough_CompactNetworkErrorsTriggerFailover(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	tests := []struct {
+		name           string
+		resp           *http.Response
+		err            error
+		expectFailover bool
+	}{
+		{
+			name:           "request_error",
+			err:            errors.New("stream disconnected before completion"),
+			expectFailover: true,
+		},
+		{
+			name: "read_error",
+			resp: &http.Response{
+				StatusCode: http.StatusOK,
+				Header:     http.Header{"Content-Type": []string{"application/json"}, "x-request-id": []string{"rid-compact"}},
+				Body:       passthroughErrReadCloser{err: io.ErrUnexpectedEOF},
+			},
+			expectFailover: false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			rec := httptest.NewRecorder()
+			c, _ := gin.CreateTestContext(rec)
+			c.Request = httptest.NewRequest(http.MethodPost, "/v1/responses/compact", bytes.NewReader(nil))
+			c.Request.Header.Set("User-Agent", "codex_cli_rs/0.1.0")
+
+			upstream := &httpUpstreamRecorder{resp: tt.resp, err: tt.err}
+			svc := &OpenAIGatewayService{
+				cfg:          &config.Config{Gateway: config.GatewayConfig{ForceCodexCLI: false}},
+				httpUpstream: upstream,
+			}
+			account := &Account{
+				ID:             123,
+				Name:           "acc",
+				Platform:       PlatformOpenAI,
+				Type:           AccountTypeOAuth,
+				Concurrency:    1,
+				Credentials:    map[string]any{"access_token": "oauth-token", "chatgpt_account_id": "chatgpt-acc"},
+				Extra:          map[string]any{"openai_passthrough": true},
+				Status:         StatusActive,
+				Schedulable:    true,
+				RateMultiplier: f64p(1),
+			}
+			body := []byte(`{"model":"gpt-5.5","instructions":"local-test-instructions","input":[{"type":"text","text":"compact me"}]}`)
+
+			_, err := svc.Forward(context.Background(), c, account, body)
+			require.Error(t, err)
+			var failoverErr *UpstreamFailoverError
+			if tt.expectFailover {
+				require.ErrorAs(t, err, &failoverErr)
+				require.Equal(t, http.StatusBadGateway, failoverErr.StatusCode)
+				require.False(t, c.Writer.Written(), "compact 网络错误应交给外层 failover，而不是直接写回客户端")
+			} else {
+				require.False(t, errors.As(err, &failoverErr))
+				require.ErrorIs(t, err, io.ErrUnexpectedEOF)
+				require.False(t, c.Writer.Written())
+			}
+		})
+	}
+}
+
 func TestOpenAIGatewayService_OAuthPassthrough_NonCodexUAFallbackToCodexUA(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
diff --git a/backend/internal/service/openai_ws_forwarder.go b/backend/internal/service/openai_ws_forwarder.go
index f57515302d..067eeb6029 100644
--- a/backend/internal/service/openai_ws_forwarder.go
+++ b/backend/internal/service/openai_ws_forwarder.go
@@ -2524,12 +2524,14 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 				hooks,
 				wsDecision,
 			)
+		case OpenAIWSIngressModeHTTPBridge:
+			forceHTTPBridge = true
 		case OpenAIWSIngressModeCtxPool, OpenAIWSIngressModeShared, OpenAIWSIngressModeDedicated:
 			// continue
 		default:
 			return NewOpenAIWSClientCloseError(
 				coderws.StatusPolicyViolation,
-				"websocket mode only supports ctx_pool/passthrough",
+				"websocket mode only supports ctx_pool/passthrough/http_bridge",
 				nil,
 			)
 		}
@@ -2840,7 +2842,7 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient(
 	}
 	refreshIngressRouteState(firstPayload)
 
-	if s.shouldBridgeOpenAIWSHTTP(account, firstPayload.payloadBytes, firstPayload.previousResponseID) {
+	if forceHTTPBridge || s.shouldBridgeOpenAIWSHTTP(account, firstPayload.payloadBytes, firstPayload.previousResponseID) {
 		logOpenAIWSModeInfo(
 			"ingress_ws_http_bridge_start account_id=%d account_type=%s payload_bytes=%d threshold_bytes=%d has_session_hash=%v store_disabled=%v",
 			account.ID,
diff --git a/backend/internal/service/openai_ws_forwarder_ingress_session_test.go b/backend/internal/service/openai_ws_forwarder_ingress_session_test.go
index 856e81be02..9ae1b855ea 100644
--- a/backend/internal/service/openai_ws_forwarder_ingress_session_test.go
+++ b/backend/internal/service/openai_ws_forwarder_ingress_session_test.go
@@ -830,6 +830,152 @@ func TestOpenAIGatewayService_ProxyResponsesWebSocketFromClient_PassthroughHeade
 	require.Equal(t, "turn-meta-1", captureDialer.lastHeaders.Get(openAIWSTurnMetadataHeader))
 }
 
+func TestOpenAIGatewayService_ProxyResponsesWebSocketFromClient_HTTPBridgeModeRelaysHTTPStream(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	cfg := &config.Config{}
+	cfg.Security.URLAllowlist.Enabled = false
+	cfg.Security.URLAllowlist.AllowInsecureHTTP = true
+	cfg.Gateway.OpenAIWS.Enabled = true
+	cfg.Gateway.OpenAIWS.OAuthEnabled = true
+	cfg.Gateway.OpenAIWS.APIKeyEnabled = true
+	cfg.Gateway.OpenAIWS.ResponsesWebsocketsV2 = true
+	cfg.Gateway.OpenAIWS.ModeRouterV2Enabled = true
+	cfg.Gateway.OpenAIWS.IngressModeDefault = OpenAIWSIngressModeCtxPool
+	cfg.Gateway.OpenAIWS.ReadTimeoutSeconds = 3
+	cfg.Gateway.OpenAIWS.WriteTimeoutSeconds = 3
+
+	upstream := &httpUpstreamRecorder{
+		resp: &http.Response{
+			StatusCode: http.StatusOK,
+			Header:     http.Header{"Content-Type": []string{"text/event-stream"}, "x-request-id": []string{"rid_bridge_1"}},
+			Body: io.NopCloser(strings.NewReader(
+				"data: {\"type\":\"response.output_text.delta\",\"delta\":\"hello\"}\n\n" +
+					"data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_http_bridge_1\",\"usage\":{\"input_tokens\":2,\"output_tokens\":1,\"input_tokens_details\":{\"cached_tokens\":1}}}}\n\n" +
+					"data: [DONE]\n\n",
+			)),
+		},
+	}
+	svc := &OpenAIGatewayService{
+		cfg:              cfg,
+		httpUpstream:     upstream,
+		cache:            &stubGatewayCache{},
+		openaiWSResolver: NewOpenAIWSProtocolResolver(cfg),
+		toolCorrector:    NewCodexToolCorrector(),
+	}
+
+	account := &Account{
+		ID:          552,
+		Name:        "openai-ingress-http-bridge",
+		Platform:    PlatformOpenAI,
+		Type:        AccountTypeAPIKey,
+		Status:      StatusActive,
+		Schedulable: true,
+		Concurrency: 1,
+		Credentials: map[string]any{
+			"api_key": "sk-test",
+		},
+		Extra: map[string]any{
+			"openai_apikey_responses_websockets_v2_mode": OpenAIWSIngressModeHTTPBridge,
+		},
+	}
+
+	serverErrCh := make(chan error, 1)
+	resultCh := make(chan *OpenAIForwardResult, 1)
+	hooks := &OpenAIWSIngressHooks{
+		AfterTurn: func(_ int, result *OpenAIForwardResult, turnErr error) {
+			if turnErr == nil && result != nil {
+				resultCh <- result
+			}
+		},
+	}
+
+	wsServer := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		conn, err := coderws.Accept(w, r, &coderws.AcceptOptions{
+			CompressionMode: coderws.CompressionContextTakeover,
+		})
+		if err != nil {
+			serverErrCh <- err
+			return
+		}
+		defer func() {
+			_ = conn.CloseNow()
+		}()
+
+		rec := httptest.NewRecorder()
+		ginCtx, _ := gin.CreateTestContext(rec)
+		req := r.Clone(r.Context())
+		req.Header = req.Header.Clone()
+		req.Header.Set("User-Agent", "unit-test-agent/1.0")
+		ginCtx.Request = req
+
+		readCtx, cancel := context.WithTimeout(r.Context(), 3*time.Second)
+		msgType, firstMessage, readErr := conn.Read(readCtx)
+		cancel()
+		if readErr != nil {
+			serverErrCh <- readErr
+			return
+		}
+		if msgType != coderws.MessageText && msgType != coderws.MessageBinary {
+			serverErrCh <- errors.New("unsupported websocket client message type")
+			return
+		}
+
+		serverErrCh <- svc.ProxyResponsesWebSocketFromClient(r.Context(), ginCtx, conn, account, "sk-test", firstMessage, hooks)
+	}))
+	defer wsServer.Close()
+
+	dialCtx, cancelDial := context.WithTimeout(context.Background(), 3*time.Second)
+	clientConn, _, err := coderws.Dial(dialCtx, "ws"+strings.TrimPrefix(wsServer.URL, "http"), nil)
+	cancelDial()
+	require.NoError(t, err)
+	defer func() {
+		_ = clientConn.CloseNow()
+	}()
+
+	writeCtx, cancelWrite := context.WithTimeout(context.Background(), 3*time.Second)
+	err = clientConn.Write(writeCtx, coderws.MessageText, []byte(`{"type":"response.create","model":"gpt-5.1","stream":false}`))
+	cancelWrite()
+	require.NoError(t, err)
+
+	readCtx, cancelRead := context.WithTimeout(context.Background(), 3*time.Second)
+	_, event1, readErr1 := clientConn.Read(readCtx)
+	cancelRead()
+	require.NoError(t, readErr1)
+	require.Equal(t, "response.output_text.delta", gjson.GetBytes(event1, "type").String())
+	require.Equal(t, "hello", gjson.GetBytes(event1, "delta").String())
+
+	readCtx2, cancelRead2 := context.WithTimeout(context.Background(), 3*time.Second)
+	_, event2, readErr2 := clientConn.Read(readCtx2)
+	cancelRead2()
+	require.NoError(t, readErr2)
+	require.Equal(t, "response.completed", gjson.GetBytes(event2, "type").String())
+	require.Equal(t, "resp_http_bridge_1", gjson.GetBytes(event2, "response.id").String())
+
+	_ = clientConn.Close(coderws.StatusNormalClosure, "done")
+
+	select {
+	case serverErr := <-serverErrCh:
+		require.NoError(t, serverErr)
+	case <-time.After(5 * time.Second):
+		t.Fatal("等待 http_bridge websocket 结束超时")
+	}
+
+	select {
+	case result := <-resultCh:
+		require.Equal(t, "resp_http_bridge_1", result.RequestID)
+		require.True(t, result.OpenAIWSMode)
+		require.Equal(t, 2, result.Usage.InputTokens)
+		require.Equal(t, 1, result.Usage.OutputTokens)
+		require.Equal(t, 1, result.Usage.CacheReadInputTokens)
+		require.NotNil(t, result.FirstTokenMs)
+	case <-time.After(2 * time.Second):
+		t.Fatal("未收到 http_bridge turn 结果回调")
+	}
+
+	require.NotNil(t, upstream.lastReq, "http_bridge 模式应调用 HTTP 上游")
+}
+
 func TestOpenAIGatewayService_ProxyResponsesWebSocketFromClient_ModeOffReturnsPolicyViolation(t *testing.T) {
 	gin.SetMode(gin.TestMode)
 
diff --git a/backend/internal/service/openai_ws_protocol_resolver.go b/backend/internal/service/openai_ws_protocol_resolver.go
index 7266759c88..e11975bf37 100644
--- a/backend/internal/service/openai_ws_protocol_resolver.go
+++ b/backend/internal/service/openai_ws_protocol_resolver.go
@@ -10,6 +10,9 @@ const (
 	OpenAIUpstreamTransportHTTPSSE              OpenAIUpstreamTransport = "http_sse"
 	OpenAIUpstreamTransportResponsesWebsocket   OpenAIUpstreamTransport = "responses_websockets"
 	OpenAIUpstreamTransportResponsesWebsocketV2 OpenAIUpstreamTransport = "responses_websockets_v2"
+	// OpenAIUpstreamTransportResponsesWebsocketV2Ingress 用于 WS ingress 入口选账号：
+	// mode_router_v2 开启时允许 ctx_pool/passthrough/http_bridge，拒绝 off。
+	OpenAIUpstreamTransportResponsesWebsocketV2Ingress OpenAIUpstreamTransport = "responses_websockets_v2_ingress"
 )
 
 // OpenAIWSProtocolDecision 表示协议决策结果。
@@ -71,6 +74,8 @@ func (r *defaultOpenAIWSProtocolResolver) Resolve(account *Account) OpenAIWSProt
 			return openAIWSHTTPDecision("account_mode_off")
 		case OpenAIWSIngressModeCtxPool, OpenAIWSIngressModePassthrough:
 			// continue
+		case OpenAIWSIngressModeHTTPBridge:
+			return openAIWSHTTPDecision("ws_v2_mode_http_bridge")
 		case OpenAIWSIngressModeShared, OpenAIWSIngressModeDedicated:
 			// 历史值兼容：按 ctx_pool 处理。
 			mode = OpenAIWSIngressModeCtxPool
diff --git a/backend/internal/service/openai_ws_protocol_resolver_test.go b/backend/internal/service/openai_ws_protocol_resolver_test.go
index 4d5dc5f18b..6d18604d03 100644
--- a/backend/internal/service/openai_ws_protocol_resolver_test.go
+++ b/backend/internal/service/openai_ws_protocol_resolver_test.go
@@ -202,6 +202,20 @@ func TestOpenAIWSProtocolResolver_Resolve_ModeRouterV2(t *testing.T) {
 		require.Equal(t, "ws_v2_mode_passthrough", decision.Reason)
 	})
 
+	t.Run("http_bridge mode routes to http_sse", func(t *testing.T) {
+		httpBridgeAccount := &Account{
+			Platform:    PlatformOpenAI,
+			Type:        AccountTypeOAuth,
+			Concurrency: 1,
+			Extra: map[string]any{
+				"openai_oauth_responses_websockets_v2_mode": OpenAIWSIngressModeHTTPBridge,
+			},
+		}
+		decision := NewOpenAIWSProtocolResolver(cfg).Resolve(httpBridgeAccount)
+		require.Equal(t, OpenAIUpstreamTransportHTTPSSE, decision.Transport)
+		require.Equal(t, "ws_v2_mode_http_bridge", decision.Reason)
+	})
+
 	t.Run("non-positive concurrency is rejected in v2 router", func(t *testing.T) {
 		invalidConcurrency := &Account{
 			Platform: PlatformOpenAI,
diff --git a/deploy/.env.example b/deploy/.env.example
index e80663ef1f..59e4b44b91 100644
--- a/deploy/.env.example
+++ b/deploy/.env.example
@@ -251,6 +251,9 @@ RATE_LIMIT_OVERLOAD_COOLDOWN_MINUTES=10
 #
 # 默认：false
 GATEWAY_FORCE_CODEX_CLI=false
+# OpenAI /responses/compact 上游模型（默认 gpt-5.4）。
+# 当 compact 端点暂未支持更新模型时，可通过这里降级规避失败。
+GATEWAY_OPENAI_COMPACT_MODEL=gpt-5.4
 # OpenAI/Codex 等待上游响应头超时（秒）；0 表示不使用本地响应头超时截断。
 GATEWAY_OPENAI_RESPONSE_HEADER_TIMEOUT=0
 # OpenAI HTTP 上游默认启用 HTTP/2；如需紧急回滚可设为 false。
diff --git a/deploy/config.example.yaml b/deploy/config.example.yaml
index 566c277196..5a3afb0315 100644
--- a/deploy/config.example.yaml
+++ b/deploy/config.example.yaml
@@ -242,11 +242,16 @@ gateway:
   # OpenAI 透传模式是否放行客户端超时头（如 x-stainless-timeout）
   # 默认 false：过滤超时头，降低上游提前断流风险。
   openai_passthrough_allow_timeout_headers: false
+  # Model used for OpenAI /responses/compact upstream requests (default: gpt-5.4).
+  # OpenAI /responses/compact 上游模型（默认 gpt-5.4）。
+  # Use this to avoid compact failures when newer models are not yet supported by the compact endpoint.
+  # 当 compact 端点暂未支持更新模型时，可通过这里降级规避失败。
+  openai_compact_model: "gpt-5.4"
   # OpenAI Responses WebSocket 配置（默认开启，可按需回滚到 HTTP）
   openai_ws:
     # 新版 WS mode 路由（默认关闭）。关闭时保持当前 legacy 实现行为。
     mode_router_v2_enabled: false
-    # ingress 默认模式：off|ctx_pool|passthrough（仅 mode_router_v2_enabled=true 生效）
+    # ingress 默认模式：off|ctx_pool|passthrough|http_bridge（仅 mode_router_v2_enabled=true 生效）
     # 兼容旧值：shared/dedicated 会按 ctx_pool 处理。
     ingress_mode_default: ctx_pool
     # 全局总开关，默认 true；关闭时所有请求保持原有 HTTP/SSE 路由
diff --git a/frontend/src/components/account/BulkEditAccountModal.vue b/frontend/src/components/account/BulkEditAccountModal.vue
index c062befebe..d24488b47a 100644
--- a/frontend/src/components/account/BulkEditAccountModal.vue
+++ b/frontend/src/components/account/BulkEditAccountModal.vue
@@ -1153,6 +1153,7 @@ import {
   OPENAI_WS_MODE_CTX_POOL,
   OPENAI_WS_MODE_OFF,
   OPENAI_WS_MODE_PASSTHROUGH,
+  OPENAI_WS_MODE_HTTP_BRIDGE,
   isOpenAIWSModeEnabled,
   resolveOpenAIWSModeConcurrencyHintKey
 } from '@/utils/openaiWsMode'
@@ -1194,7 +1195,7 @@ const allOpenAIPassthroughCapable = computed(() => {
     targetSelectedPlatforms.value.length === 1 &&
     targetSelectedPlatforms.value[0] === 'openai' &&
     targetSelectedTypes.value.length > 0 &&
-    targetSelectedTypes.value.every(t => t === 'oauth' || t === 'apikey')
+    targetSelectedTypes.value.every(t => t === 'oauth' || t === 'setup-token' || t === 'apikey')
   )
 })
 
@@ -1203,7 +1204,7 @@ const allOpenAIOAuth = computed(() => {
     targetSelectedPlatforms.value.length === 1 &&
     targetSelectedPlatforms.value[0] === 'openai' &&
     targetSelectedTypes.value.length > 0 &&
-    targetSelectedTypes.value.every(t => t === 'oauth')
+    targetSelectedTypes.value.every(t => t === 'oauth' || t === 'setup-token')
   )
 })
 
@@ -1330,7 +1331,8 @@ const isOpenAIModelRestrictionDisabled = computed(
 const openAIWSModeOptions = computed(() => [
   { value: OPENAI_WS_MODE_OFF, label: t('admin.accounts.openai.wsModeOff') },
   { value: OPENAI_WS_MODE_CTX_POOL, label: t('admin.accounts.openai.wsModeCtxPool') },
-  { value: OPENAI_WS_MODE_PASSTHROUGH, label: t('admin.accounts.openai.wsModePassthrough') }
+  { value: OPENAI_WS_MODE_PASSTHROUGH, label: t('admin.accounts.openai.wsModePassthrough') },
+  { value: OPENAI_WS_MODE_HTTP_BRIDGE, label: t('admin.accounts.openai.wsModeHttpBridge') }
 ])
 const openAICompactModeOptions = computed(() => [
   { value: 'auto', label: t('admin.accounts.openai.compactModeAuto') },
diff --git a/frontend/src/components/account/CreateAccountModal.vue b/frontend/src/components/account/CreateAccountModal.vue
index c43709a7ac..3514153c82 100644
--- a/frontend/src/components/account/CreateAccountModal.vue
+++ b/frontend/src/components/account/CreateAccountModal.vue
@@ -3337,6 +3337,7 @@ import {
   OPENAI_WS_MODE_CTX_POOL,
   OPENAI_WS_MODE_OFF,
   OPENAI_WS_MODE_PASSTHROUGH,
+  OPENAI_WS_MODE_HTTP_BRIDGE,
   isOpenAIWSModeEnabled,
   resolveOpenAIWSModeConcurrencyHintKey,
   type OpenAIWSMode
@@ -3703,7 +3704,8 @@ const geminiSelectedTier = computed(() => {
 const openAIWSModeOptions = computed(() => [
   { value: OPENAI_WS_MODE_OFF, label: t('admin.accounts.openai.wsModeOff') },
   { value: OPENAI_WS_MODE_CTX_POOL, label: t('admin.accounts.openai.wsModeCtxPool') },
-  { value: OPENAI_WS_MODE_PASSTHROUGH, label: t('admin.accounts.openai.wsModePassthrough') }
+  { value: OPENAI_WS_MODE_PASSTHROUGH, label: t('admin.accounts.openai.wsModePassthrough') },
+  { value: OPENAI_WS_MODE_HTTP_BRIDGE, label: t('admin.accounts.openai.wsModeHttpBridge') }
 ])
 
 const openaiResponsesWebSocketV2Mode = computed({
diff --git a/frontend/src/components/account/EditAccountModal.vue b/frontend/src/components/account/EditAccountModal.vue
index 9a815ee813..b5f97753d8 100644
--- a/frontend/src/components/account/EditAccountModal.vue
+++ b/frontend/src/components/account/EditAccountModal.vue
@@ -1342,7 +1342,7 @@
 
       <!-- OpenAI 自动透传开关（OAuth/API Key） -->
       <div
-        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'apikey')"
+        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'setup-token' || account?.type === 'apikey')"
         class="border-t border-gray-200 pt-4 dark:border-dark-600"
       >
         <div class="flex items-center justify-between">
@@ -1372,7 +1372,7 @@
 
       <!-- OpenAI Codex 图片生成桥接账号级覆盖 -->
       <div
-        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'apikey')"
+        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'setup-token' || account?.type === 'apikey')"
         class="border-t border-gray-200 pt-4 dark:border-dark-600"
       >
         <div class="overflow-hidden rounded-lg border border-sky-100 bg-sky-50/60 shadow-sm dark:border-sky-900/50 dark:bg-sky-950/20">
@@ -1432,7 +1432,7 @@
 
       <!-- OpenAI WS Mode 三态（off/ctx_pool/passthrough） -->
       <div
-        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'apikey')"
+        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'setup-token' || account?.type === 'apikey')"
         class="border-t border-gray-200 pt-4 dark:border-dark-600"
       >
         <div class="flex items-center justify-between">
@@ -1446,7 +1446,7 @@
             </p>
           </div>
           <div class="w-52">
-            <Select v-model="openaiResponsesWebSocketV2Mode" :options="openAIWSModeOptions" />
+            <Select v-model="openaiResponsesWebSocketV2Mode" data-testid="edit-openai-ws-mode-select" :options="openAIWSModeOptions" />
           </div>
         </div>
       </div>
@@ -1680,7 +1680,7 @@
 
       <!-- OpenAI OAuth Codex 官方客户端限制开关 -->
       <div
-        v-if="account?.platform === 'openai' && account?.type === 'oauth'"
+        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'setup-token')"
         class="border-t border-gray-200 pt-4 dark:border-dark-600"
       >
         <div class="flex items-center justify-between">
@@ -1735,7 +1735,7 @@
       </div>
 
       <div
-        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'apikey')"
+        v-if="account?.platform === 'openai' && (account?.type === 'oauth' || account?.type === 'setup-token' || account?.type === 'apikey')"
         class="border-t border-gray-200 pt-4 dark:border-dark-600 space-y-4"
       >
         <div class="flex items-center justify-between">
@@ -2442,6 +2442,7 @@ import {
   OPENAI_WS_MODE_CTX_POOL,
   OPENAI_WS_MODE_OFF,
   OPENAI_WS_MODE_PASSTHROUGH,
+  OPENAI_WS_MODE_HTTP_BRIDGE,
   isOpenAIWSModeEnabled,
   resolveOpenAIWSModeConcurrencyHintKey,
   type OpenAIWSMode,
@@ -2660,7 +2661,8 @@ const editResetTimezone = ref<string | null>(null)
 const openAIWSModeOptions = computed(() => [
   { value: OPENAI_WS_MODE_OFF, label: t('admin.accounts.openai.wsModeOff') },
   { value: OPENAI_WS_MODE_CTX_POOL, label: t('admin.accounts.openai.wsModeCtxPool') },
-  { value: OPENAI_WS_MODE_PASSTHROUGH, label: t('admin.accounts.openai.wsModePassthrough') }
+  { value: OPENAI_WS_MODE_PASSTHROUGH, label: t('admin.accounts.openai.wsModePassthrough') },
+  { value: OPENAI_WS_MODE_HTTP_BRIDGE, label: t('admin.accounts.openai.wsModeHttpBridge') }
 ])
 const openaiResponsesWebSocketV2Mode = computed({
   get: () => {
@@ -3023,7 +3025,7 @@ const syncFormFromAccount = (newAccount: Account | null) => {
 	autoPause5hDisabled.value = extra?.auto_pause_5h_disabled === true
 	autoPause7dDisabled.value = extra?.auto_pause_7d_disabled === true
 
-  // Load OpenAI passthrough toggle (OpenAI OAuth/API Key)
+  // Load OpenAI passthrough toggle (OpenAI OAuth/SetupToken/API Key)
   openaiPassthroughEnabled.value = false
   openAICompactMode.value = 'auto'
   openAIResponsesMode.value = 'auto'
@@ -3037,7 +3039,7 @@ const syncFormFromAccount = (newAccount: Account | null) => {
   anthropicPassthroughEnabled.value = false
   anthropicAPIKeyAuthScheme.value = 'x_api_key'
   webSearchEmulationMode.value = 'default'
-  if (newAccount.platform === 'openai' && (newAccount.type === 'oauth' || newAccount.type === 'apikey')) {
+  if (newAccount.platform === 'openai' && (newAccount.type === 'oauth' || newAccount.type === 'setup-token' || newAccount.type === 'apikey')) {
     openaiPassthroughEnabled.value = extra?.openai_passthrough === true || extra?.openai_oauth_passthrough === true
     openAICompactMode.value = (extra?.openai_compact_mode as OpenAICompactMode) || 'auto'
     if (newAccount.type === 'apikey') {
@@ -3069,7 +3071,7 @@ const syncFormFromAccount = (newAccount: Account | null) => {
       fallbackEnabledKeys: ['responses_websockets_v2_enabled', 'openai_ws_enabled'],
       defaultMode: OPENAI_WS_MODE_OFF
     })
-    if (newAccount.type === 'oauth') {
+    if (newAccount.type === 'oauth' || newAccount.type === 'setup-token') {
       codexCLIOnlyEnabled.value = extra?.codex_cli_only === true
       codexCLIOnlyAppServerEnabled.value =
         extra?.codex_cli_only_allow_app_server === true
@@ -4138,12 +4140,12 @@ const handleSubmit = async () => {
       updatePayload.extra = newExtra
     }
 
-    // For OpenAI OAuth/API Key accounts, handle passthrough mode in extra
-	if (props.account.platform === 'openai' && (props.account.type === 'oauth' || props.account.type === 'apikey')) {
-		const currentExtra = (props.account.extra as Record<string, unknown>) || {}
-		const newExtra: Record<string, unknown> = { ...currentExtra }
+    // For OpenAI OAuth/SetupToken/API Key accounts, handle passthrough mode in extra
+    if (props.account.platform === 'openai' && (props.account.type === 'oauth' || props.account.type === 'setup-token' || props.account.type === 'apikey')) {
+      const currentExtra = (props.account.extra as Record<string, unknown>) || {}
+      const newExtra: Record<string, unknown> = { ...currentExtra }
       const hadCodexCLIOnlyEnabled = currentExtra.codex_cli_only === true
-      if (props.account.type === 'oauth') {
+      if (props.account.type === 'oauth' || props.account.type === 'setup-token') {
         newExtra.openai_oauth_responses_websockets_v2_mode = openaiOAuthResponsesWebSocketV2Mode.value
         newExtra.openai_oauth_responses_websockets_v2_enabled = isOpenAIWSModeEnabled(openaiOAuthResponsesWebSocketV2Mode.value)
       } else if (props.account.type === 'apikey') {
@@ -4198,7 +4200,7 @@ const handleSubmit = async () => {
         newExtra.codex_image_generation_bridge = codexImageGenerationBridgeMode.value === 'enabled'
       }
 
-      if (props.account.type === 'oauth') {
+      if (props.account.type === 'oauth' || props.account.type === 'setup-token') {
         if (codexCLIOnlyEnabled.value) {
           newExtra.codex_cli_only = true
         } else if (hadCodexCLIOnlyEnabled) {
diff --git a/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts b/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts
index 1ccf8ed363..31f6e3bd26 100644
--- a/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts
+++ b/frontend/src/components/account/__tests__/BulkEditAccountModal.spec.ts
@@ -149,21 +149,21 @@ describe('BulkEditAccountModal', () => {
     })
   })
 
-  it('OpenAI OAuth 批量编辑应提交 OAuth 专属 WS mode 字段', async () => {
+  it('OpenAI OAuth 批量编辑应提交 OAuth 专属 WS mode 字段（含 http_bridge）', async () => {
     const wrapper = mountModal({
       selectedPlatforms: ['openai'],
       selectedTypes: ['oauth']
     })
 
     await wrapper.get('#bulk-edit-openai-ws-mode-enabled').setValue(true)
-    await wrapper.get('[data-testid="bulk-edit-openai-ws-mode-select"]').setValue('passthrough')
+    await wrapper.get('[data-testid="bulk-edit-openai-ws-mode-select"]').setValue('http_bridge')
     await wrapper.get('#bulk-edit-account-form').trigger('submit.prevent')
     await flushPromises()
 
     expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledTimes(1)
     expect(adminAPI.accounts.bulkUpdate).toHaveBeenCalledWith([1, 2], {
       extra: {
-        openai_oauth_responses_websockets_v2_mode: 'passthrough',
+        openai_oauth_responses_websockets_v2_mode: 'http_bridge',
         openai_oauth_responses_websockets_v2_enabled: true
       }
     })
diff --git a/frontend/src/components/account/__tests__/EditAccountModal.spec.ts b/frontend/src/components/account/__tests__/EditAccountModal.spec.ts
index de59b4d5fe..381d512ee7 100644
--- a/frontend/src/components/account/__tests__/EditAccountModal.spec.ts
+++ b/frontend/src/components/account/__tests__/EditAccountModal.spec.ts
@@ -241,6 +241,17 @@ function buildAntigravityAccount(projectId = 'configured-project') {
   } as any
 }
 
+function buildOpenAISetupTokenAccount() {
+  return {
+    ...buildAccount(),
+    type: 'setup-token',
+    extra: {
+      openai_oauth_responses_websockets_v2_mode: 'ctx_pool',
+      openai_oauth_responses_websockets_v2_enabled: true
+    }
+  } as any
+}
+
 function mountModal(account = buildAccount()) {
   return mount(EditAccountModal, {
     props: {
@@ -566,6 +577,23 @@ describe('EditAccountModal', () => {
     expect(updateAccountMock.mock.calls[0]?.[1]?.extra).not.toHaveProperty('codex_image_generation_bridge_enabled')
   })
 
+  it('setup-token account can select and submit OAuth WS mode', async () => {
+    const account = buildOpenAISetupTokenAccount()
+    updateAccountMock.mockReset()
+    checkMixedChannelRiskMock.mockReset()
+    checkMixedChannelRiskMock.mockResolvedValue({ has_risk: false })
+    updateAccountMock.mockResolvedValue(account)
+
+    const wrapper = mountModal(account)
+
+    await wrapper.get('[data-testid="edit-openai-ws-mode-select"]').setValue('http_bridge')
+    await wrapper.get('form#edit-account-form').trigger('submit.prevent')
+
+    expect(updateAccountMock).toHaveBeenCalledTimes(1)
+    expect(updateAccountMock.mock.calls[0]?.[1]?.extra?.openai_oauth_responses_websockets_v2_mode).toBe('http_bridge')
+    expect(updateAccountMock.mock.calls[0]?.[1]?.extra?.openai_oauth_responses_websockets_v2_enabled).toBe(true)
+  })
+
   it('allows saving apikey account when backend redacted api_key but credentials_status reports it exists', async () => {
     // 新前端 + 新后端：响应已脱敏，credentials 里没有 api_key，credentials_status.has_api_key=true
     const account = buildAccount()
diff --git a/frontend/src/i18n/locales/en.ts b/frontend/src/i18n/locales/en.ts
index cefd25279f..831b7d7eae 100644
--- a/frontend/src/i18n/locales/en.ts
+++ b/frontend/src/i18n/locales/en.ts
@@ -3515,6 +3515,7 @@ export default {
         wsModeOff: 'Off (off)',
         wsModeCtxPool: 'Context Pool (ctx_pool)',
         wsModePassthrough: 'Passthrough (passthrough)',
+        wsModeHttpBridge: 'HTTP Bridge (http_bridge)',
         wsModeShared: 'Shared (shared)',
         wsModeDedicated: 'Dedicated (dedicated)',
         wsModeConcurrencyHint:
diff --git a/frontend/src/i18n/locales/zh.ts b/frontend/src/i18n/locales/zh.ts
index d65d9250c0..64c62f39f8 100644
--- a/frontend/src/i18n/locales/zh.ts
+++ b/frontend/src/i18n/locales/zh.ts
@@ -3693,6 +3693,7 @@ export default {
         wsModeOff: '关闭（off）',
         wsModeCtxPool: '上下文池（ctx_pool）',
         wsModePassthrough: '透传（passthrough）',
+        wsModeHttpBridge: 'HTTP 桥接（http_bridge）',
         wsModeShared: '共享（shared）',
         wsModeDedicated: '独享（dedicated）',
         wsModeConcurrencyHint: '启用 WS mode 后，该账号并发数将作为该账号 WS 连接池上限。',
diff --git a/frontend/src/utils/__tests__/openaiWsMode.spec.ts b/frontend/src/utils/__tests__/openaiWsMode.spec.ts
index 8e4f33b2cc..5555224b27 100644
--- a/frontend/src/utils/__tests__/openaiWsMode.spec.ts
+++ b/frontend/src/utils/__tests__/openaiWsMode.spec.ts
@@ -1,6 +1,7 @@
 import { describe, expect, it } from 'vitest'
 import {
   OPENAI_WS_MODE_CTX_POOL,
+  OPENAI_WS_MODE_HTTP_BRIDGE,
   OPENAI_WS_MODE_OFF,
   OPENAI_WS_MODE_PASSTHROUGH,
   isOpenAIWSModeEnabled,
@@ -15,6 +16,7 @@ describe('openaiWsMode utils', () => {
     expect(normalizeOpenAIWSMode('off')).toBe(OPENAI_WS_MODE_OFF)
     expect(normalizeOpenAIWSMode('ctx_pool')).toBe(OPENAI_WS_MODE_CTX_POOL)
     expect(normalizeOpenAIWSMode('passthrough')).toBe(OPENAI_WS_MODE_PASSTHROUGH)
+    expect(normalizeOpenAIWSMode('http_bridge')).toBe(OPENAI_WS_MODE_HTTP_BRIDGE)
     expect(normalizeOpenAIWSMode(' Shared ')).toBe(OPENAI_WS_MODE_CTX_POOL)
     expect(normalizeOpenAIWSMode('DEDICATED')).toBe(OPENAI_WS_MODE_CTX_POOL)
     expect(normalizeOpenAIWSMode('invalid')).toBeNull()
@@ -54,6 +56,7 @@ describe('openaiWsMode utils', () => {
     expect(isOpenAIWSModeEnabled(OPENAI_WS_MODE_OFF)).toBe(false)
     expect(isOpenAIWSModeEnabled(OPENAI_WS_MODE_CTX_POOL)).toBe(true)
     expect(isOpenAIWSModeEnabled(OPENAI_WS_MODE_PASSTHROUGH)).toBe(true)
+    expect(isOpenAIWSModeEnabled(OPENAI_WS_MODE_HTTP_BRIDGE)).toBe(true)
   })
 
   it('resolves concurrency hint key by mode', () => {
@@ -66,5 +69,8 @@ describe('openaiWsMode utils', () => {
     expect(resolveOpenAIWSModeConcurrencyHintKey(OPENAI_WS_MODE_PASSTHROUGH)).toBe(
       'admin.accounts.openai.wsModePassthroughHint'
     )
+    expect(resolveOpenAIWSModeConcurrencyHintKey(OPENAI_WS_MODE_HTTP_BRIDGE)).toBe(
+      'admin.accounts.openai.wsModePassthroughHint'
+    )
   })
 })
diff --git a/frontend/src/utils/openaiWsMode.ts b/frontend/src/utils/openaiWsMode.ts
index 52eba8b033..77872bc656 100644
--- a/frontend/src/utils/openaiWsMode.ts
+++ b/frontend/src/utils/openaiWsMode.ts
@@ -1,16 +1,19 @@
 export const OPENAI_WS_MODE_OFF = 'off'
 export const OPENAI_WS_MODE_CTX_POOL = 'ctx_pool'
 export const OPENAI_WS_MODE_PASSTHROUGH = 'passthrough'
+export const OPENAI_WS_MODE_HTTP_BRIDGE = 'http_bridge'
 
 export type OpenAIWSMode =
   | typeof OPENAI_WS_MODE_OFF
   | typeof OPENAI_WS_MODE_CTX_POOL
   | typeof OPENAI_WS_MODE_PASSTHROUGH
+  | typeof OPENAI_WS_MODE_HTTP_BRIDGE
 
 const OPENAI_WS_MODES = new Set<OpenAIWSMode>([
   OPENAI_WS_MODE_OFF,
   OPENAI_WS_MODE_CTX_POOL,
-  OPENAI_WS_MODE_PASSTHROUGH
+  OPENAI_WS_MODE_PASSTHROUGH,
+  OPENAI_WS_MODE_HTTP_BRIDGE
 ])
 
 export interface ResolveOpenAIWSModeOptions {
@@ -44,7 +47,7 @@ export const isOpenAIWSModeEnabled = (mode: OpenAIWSMode): boolean => {
 export const resolveOpenAIWSModeConcurrencyHintKey = (
   mode: OpenAIWSMode
 ): 'admin.accounts.openai.wsModeConcurrencyHint' | 'admin.accounts.openai.wsModePassthroughHint' => {
-  if (mode === OPENAI_WS_MODE_PASSTHROUGH) {
+  if (mode === OPENAI_WS_MODE_PASSTHROUGH || mode === OPENAI_WS_MODE_HTTP_BRIDGE) {
     return 'admin.accounts.openai.wsModePassthroughHint'
   }
   return 'admin.accounts.openai.wsModeConcurrencyHint'