feat(codex): support ultrafast service tier

Accept service_tier=ultrafast on the OpenAI/Codex gateway, advertise it
on GPT-5.6 Sol in the Codex model catalog, and include it in ChatGPT
OAuth routing hints so Codex clients can select Ultrafast.

backend/internal/handler/admin/admin_helpers_test.go

@@ -278,13 +278,15 @@ func TestOpenAIFastPolicySettingsFromDTO_NormalizesServiceTier(t *testing.T) {
 			Rules: []dto.OpenAIFastPolicyRule{
 				{ServiceTier: "priority", Action: "filter", Scope: "all"},
 				{ServiceTier: "flex", Action: "block", Scope: "oauth"},
+				{ServiceTier: "ultrafast", Action: "pass", Scope: "all"},
 				{ServiceTier: "all", Action: "pass", Scope: "apikey"},
 			},
 		}
 		out := openaiFastPolicySettingsFromDTO(in)
-		require.Len(t, out.Rules, 3)
+		require.Len(t, out.Rules, 4)
 		require.Equal(t, service.OpenAIFastTierPriority, out.Rules[0].ServiceTier)
 		require.Equal(t, service.OpenAIFastTierFlex, out.Rules[1].ServiceTier)
-		require.Equal(t, service.OpenAIFastTierAny, out.Rules[2].ServiceTier)
+		require.Equal(t, service.OpenAIFastTierUltrafast, out.Rules[2].ServiceTier)
+		require.Equal(t, service.OpenAIFastTierAny, out.Rules[3].ServiceTier)
 	})
 }

backend/internal/handler/openai_service_tier_validation_test.go

@@ -17,7 +17,7 @@ import (
 // 上以 OpenAI 兼容错误结构返回 HTTP 400。这些用例在 handler 的 service_tier
 // 校验处短路，不会进入账号选择/重试。
 //
-// 合法值（fast/priority/flex/auto/default/scale）与省略/null 的接受语义由
+// 合法值（fast/priority/flex/auto/default/scale/ultrafast）与省略/null 的接受语义由
 // service 层纯校验函数 TestValidateOpenAIServiceTierField 覆盖，避免 handler
 // 测试走入真实账号选择/重试路径。
 

backend/internal/service/billing_service.go

@@ -137,7 +137,7 @@ func usePriorityServiceTierPricing(serviceTier string, pricing *ModelPricing) bo
 
 func serviceTierCostMultiplier(serviceTier string) float64 {
 	switch normalizeBillingServiceTier(serviceTier) {
-	case "priority", "fast":
+	case "priority", "fast", OpenAIFastTierUltrafast:
 		return 2.0
 	case "flex":
 		return 0.5

backend/internal/service/billing_service_test.go

@@ -1431,6 +1431,7 @@ func TestCalculateCost_LargeTokenCount(t *testing.T) {
 func TestServiceTierCostMultiplier(t *testing.T) {
 	require.InDelta(t, 2.0, serviceTierCostMultiplier("priority"), 1e-12)
 	require.InDelta(t, 2.0, serviceTierCostMultiplier(" Priority "), 1e-12)
+	require.InDelta(t, 2.0, serviceTierCostMultiplier("ultrafast"), 1e-12)
 	require.InDelta(t, 0.5, serviceTierCostMultiplier("flex"), 1e-12)
 	require.InDelta(t, 1.0, serviceTierCostMultiplier(""), 1e-12)
 	require.InDelta(t, 1.0, serviceTierCostMultiplier("default"), 1e-12)

backend/internal/service/channel_pricing_multipliers_test.go

@@ -19,6 +19,8 @@ func TestConfiguredServiceTierMultiplier(t *testing.T) {
 		{name: "priority alias", serviceTier: "priority", pricing: &ModelPricing{FastMultiplier: pricingMultiplier(2)}, want: 2},
 		{name: "flex configured", serviceTier: "flex", pricing: &ModelPricing{FlexMultiplier: pricingMultiplier(0.4)}, want: 0.4},
 		{name: "legacy fast default", serviceTier: "fast", pricing: &ModelPricing{}, want: 2},
+		{name: "ultrafast default", serviceTier: "ultrafast", pricing: &ModelPricing{}, want: 2},
+		{name: "ultrafast ignores fast multiplier", serviceTier: "ultrafast", pricing: &ModelPricing{FastMultiplier: pricingMultiplier(2.5)}, want: 2},
 		{name: "legacy flex default", serviceTier: "flex", pricing: &ModelPricing{}, want: 0.5},
 	}
 

backend/internal/service/openai_codex_models_service.go

@@ -469,15 +469,7 @@ func newConfiguredCodexModelDescriptor(modelID string) configuredCodexModelDescr
 		descriptor.DisplayName = openaiCodexDisplayName(modelID)
 		descriptor.Description = "OpenAI GPT coding model routed through Sub2API."
 		descriptor.SupportsParallelToolCalls = true
-		if configuredCodexSupportsPriorityServiceTier(modelID) {
-			descriptor.ServiceTiers = []configuredCodexServiceTier{
-				{
-					ID:          "priority",
-					Name:        "Fast",
-					Description: "Priority processing for lower latency.",
-				},
-			}
-		}
+		descriptor.ServiceTiers = configuredCodexServiceTiersForModel(modelID)
 		if isOpenAICodexReasoningGPTModel(modelID) {
 			defaultReasoningLevel := "medium"
 			if getNormalizedCodexModel(modelID) == "gpt-5.6-sol" {
@@ -501,6 +493,25 @@ func newConfiguredCodexModelDescriptor(modelID string) configuredCodexModelDescr
 	return descriptor
 }
 
+func configuredCodexServiceTiersForModel(modelID string) []configuredCodexServiceTier {
+	tiers := make([]configuredCodexServiceTier, 0, 2)
+	if configuredCodexSupportsPriorityServiceTier(modelID) {
+		tiers = append(tiers, configuredCodexServiceTier{
+			ID:          OpenAIFastTierPriority,
+			Name:        "Fast",
+			Description: "Priority processing for lower latency.",
+		})
+	}
+	if configuredCodexSupportsUltrafastServiceTier(modelID) {
+		tiers = append(tiers, configuredCodexServiceTier{
+			ID:          OpenAIFastTierUltrafast,
+			Name:        "Ultrafast",
+			Description: "Ultra-low latency processing.",
+		})
+	}
+	return tiers
+}
+
 func configuredCodexSupportsPriorityServiceTier(modelID string) bool {
 	normalized := canonicalizeOpenAIModelAliasSpelling(modelID)
 	for _, family := range []string{"gpt-5.4", "gpt-5.5", "gpt-5.6"} {
@@ -511,6 +522,10 @@ func configuredCodexSupportsPriorityServiceTier(modelID string) bool {
 	return false
 }
 
+func configuredCodexSupportsUltrafastServiceTier(modelID string) bool {
+	return normalizeKnownOpenAICodexModel(modelID) == "gpt-5.6-sol"
+}
+
 func configuredCodexGrokReasoningLevels(modelID string) []configuredCodexReasoningLevel {
 	levels := []configuredCodexReasoningLevel{
 		{Effort: "low", Description: "Fast responses with lighter reasoning"},

backend/internal/service/openai_codex_models_service_test.go

@@ -434,7 +434,7 @@ func TestBuildCodexModelsManifestAdvertisesPriorityServiceTierForFastGPTModels(t
 	body, err := BuildCodexModelsManifest([]string{
 		"gpt-5.4-mini",
 		"gpt-5.5",
-		"gpt-5.6-sol",
+		"gpt-5.6-terra",
 	})
 	require.NoError(t, err)
 	models := decodeCodexManifestModels(t, body)
@@ -452,6 +452,36 @@ func TestBuildCodexModelsManifestAdvertisesPriorityServiceTierForFastGPTModels(t
 	}
 }
 
+// Scenario: GPT-5.6 Sol 在 Fast 之外额外声明 ultrafast service tier。
+func TestBuildCodexModelsManifestAdvertisesUltrafastServiceTierForSol(t *testing.T) {
+	t.Parallel()
+
+	body, err := BuildCodexModelsManifest([]string{
+		"gpt-5.6-sol",
+		"gpt-5.6",
+	})
+	require.NoError(t, err)
+	models := decodeCodexManifestModels(t, body)
+	require.Len(t, models, 2)
+
+	wantTiers := []any{
+		map[string]any{
+			"id":          "priority",
+			"name":        "Fast",
+			"description": "Priority processing for lower latency.",
+		},
+		map[string]any{
+			"id":          "ultrafast",
+			"name":        "Ultrafast",
+			"description": "Ultra-low latency processing.",
+		},
+	}
+	for _, model := range models {
+		require.Equal(t, wantTiers, model["service_tiers"])
+		require.Nil(t, model["default_service_tier"])
+	}
+}
+
 // Scenario: 未明确支持 Fast 的型号不推测 service tier。
 func TestBuildCodexModelsManifestLeavesServiceTiersEmptyForOtherModels(t *testing.T) {
 	t.Parallel()

backend/internal/service/openai_fast_policy_test.go

@@ -100,6 +100,9 @@ func TestEvaluateOpenAIFastPolicy_DefaultPassesKnownTiers(t *testing.T) {
 	action, _ = svc.evaluateOpenAIFastPolicy(context.Background(), account, "gpt-5.5", OpenAIFastTierFlex)
 	require.Equal(t, BetaPolicyActionPass, action)
 
+	action, _ = svc.evaluateOpenAIFastPolicy(context.Background(), account, "gpt-5.6-sol", OpenAIFastTierUltrafast)
+	require.Equal(t, BetaPolicyActionPass, action)
+
 	// empty tier → pass
 	action, _ = svc.evaluateOpenAIFastPolicy(context.Background(), account, "gpt-5.5", "")
 	require.Equal(t, BetaPolicyActionPass, action)
@@ -248,6 +251,16 @@ func TestApplyOpenAIFastPolicyToBody_UserScopedRuleOverridesGlobalRule(t *testin
 	require.NotContains(t, string(updated), `"service_tier"`)
 }
 
+func TestApplyOpenAIFastPolicyToBody_PriorityFilterLeavesUltrafast(t *testing.T) {
+	svc := newOpenAIGatewayServiceWithSettings(t, openAIFastFilterPriorityPolicy())
+	account := &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
+	body := []byte(`{"model":"gpt-5.6-sol","service_tier":"ultrafast"}`)
+
+	updated, err := svc.applyOpenAIFastPolicyToBody(context.Background(), account, "gpt-5.6-sol", body)
+	require.NoError(t, err)
+	require.Equal(t, OpenAIFastTierUltrafast, gjson.GetBytes(updated, "service_tier").String())
+}
+
 func TestApplyOpenAIFastPolicyToBody_ForcePriorityRewritesKnownTier(t *testing.T) {
 	settings := &OpenAIFastPolicySettings{
 		Rules: []OpenAIFastPolicyRule{{
@@ -259,7 +272,7 @@ func TestApplyOpenAIFastPolicyToBody_ForcePriorityRewritesKnownTier(t *testing.T
 	svc := newOpenAIGatewayServiceWithSettings(t, settings)
 	account := &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
 
-	for _, tier := range []string{"flex", "auto", "default", "scale", "fast", "priority"} {
+	for _, tier := range []string{"flex", "auto", "default", "scale", "fast", "priority", "ultrafast"} {
 		body := []byte(`{"model":"gpt-5.5","service_tier":"` + tier + `"}`)
 		updated, err := svc.applyOpenAIFastPolicyToBody(context.Background(), account, "gpt-5.5", body)
 		require.NoError(t, err)
@@ -480,19 +493,28 @@ func TestSetOpenAIFastPolicySettings_Validation(t *testing.T) {
 
 	// Valid settings persisted
 	err = svc.SetOpenAIFastPolicySettings(context.Background(), &OpenAIFastPolicySettings{
-		Rules: []OpenAIFastPolicyRule{{
-			ServiceTier: OpenAIFastTierPriority,
-			Action:      OpenAIFastPolicyActionForcePriority,
-			Scope:       BetaPolicyScopeAll,
-			UserIDs:     []int64{42, 43},
-		}},
+		Rules: []OpenAIFastPolicyRule{
+			{
+				ServiceTier: OpenAIFastTierPriority,
+				Action:      OpenAIFastPolicyActionForcePriority,
+				Scope:       BetaPolicyScopeAll,
+				UserIDs:     []int64{42, 43},
+			},
+			{
+				ServiceTier: OpenAIFastTierUltrafast,
+				Action:      BetaPolicyActionPass,
+				Scope:       BetaPolicyScopeAll,
+			},
+		},
 	})
 	require.NoError(t, err)
 
 	got, err := svc.GetOpenAIFastPolicySettings(context.Background())
 	require.NoError(t, err)
-	require.Len(t, got.Rules, 1)
+	require.Len(t, got.Rules, 2)
 	require.Equal(t, OpenAIFastTierPriority, got.Rules[0].ServiceTier)
 	require.Equal(t, OpenAIFastPolicyActionForcePriority, got.Rules[0].Action)
 	require.Equal(t, []int64{42, 43}, got.Rules[0].UserIDs)
+	require.Equal(t, OpenAIFastTierUltrafast, got.Rules[1].ServiceTier)
+	require.Equal(t, BetaPolicyActionPass, got.Rules[1].Action)
 }

backend/internal/service/openai_fast_policy_ws_test.go

@@ -112,7 +112,7 @@ func TestWSResponseCreate_ForcePriorityRewritesKnownTier(t *testing.T) {
 	svc := newOpenAIGatewayServiceWithSettings(t, settings)
 	account := &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}
 
-	for _, tier := range []string{"flex", "auto", "default", "scale", "fast", "priority"} {
+	for _, tier := range []string{"flex", "auto", "default", "scale", "fast", "priority", "ultrafast"} {
 		frame := []byte(`{"type":"response.create","model":"gpt-5.5","service_tier":"` + tier + `"}`)
 		updated, blocked, err := svc.applyOpenAIFastPolicyToWSResponseCreate(context.Background(), account, "gpt-5.5", frame)
 		require.NoError(t, err)

backend/internal/service/openai_fast_service_tier_test.go

@@ -43,7 +43,7 @@ func TestValidateOpenAIServiceTierField(t *testing.T) {
 	})
 
 	t.Run("official tiers pass through", func(t *testing.T) {
-		for _, tier := range []string{"flex", "auto", "default", "scale"} {
+		for _, tier := range []string{"flex", "auto", "default", "scale", "ultrafast"} {
 			norm, err := ValidateOpenAIServiceTierField([]byte(`{"model":"gpt-5.5","service_tier":"` + tier + `"}`))
 			require.NoError(t, err, "tier %q must be accepted", tier)
 			require.Equal(t, tier, norm)

backend/internal/service/openai_gateway_chat_completions_test.go

@@ -148,6 +148,11 @@ func TestNormalizeResponsesBodyServiceTier(t *testing.T) {
 	require.Equal(t, "scale", tier)
 	require.Equal(t, "scale", gjson.GetBytes(body, "service_tier").String())
 
+	body, tier, err = normalizeResponsesBodyServiceTier([]byte(`{"model":"gpt-5.6-sol","service_tier":"ultrafast"}`))
+	require.NoError(t, err)
+	require.Equal(t, "ultrafast", tier)
+	require.Equal(t, "ultrafast", gjson.GetBytes(body, "service_tier").String())
+
 	// 真未知值才会被删除。
 	body, tier, err = normalizeResponsesBodyServiceTier([]byte(`{"model":"gpt-5.1","service_tier":"turbo"}`))
 	require.NoError(t, err)

backend/internal/service/openai_gateway_record_usage_test.go

@@ -1466,9 +1466,8 @@ func TestNormalizeOpenAIServiceTier(t *testing.T) {
 
 	t.Run("openai official tiers preserved", func(t *testing.T) {
 		// OpenAI 官方文档定义的合法 tier 值都应被透传保留，避免因白名单过窄
-		// 静默剥离客户端显式发送的合法字段。Codex 客户端只发 priority/flex，
-		// 所以扩大白名单对 Codex 流量零影响（见 codex-rs/core/src/client.rs）。
-		for _, tier := range []string{"priority", "flex", "auto", "default", "scale"} {
+		// 静默剥离客户端显式发送的合法字段。Codex 会发 priority/flex/ultrafast。
+		for _, tier := range []string{"priority", "flex", "auto", "default", "scale", "ultrafast"} {
 			got := normalizeOpenAIServiceTier(tier)
 			require.NotNil(t, got, "tier %q should not be normalized to nil", tier)
 			require.Equal(t, tier, *got)
@@ -1487,6 +1486,7 @@ func TestExtractOpenAIServiceTier(t *testing.T) {
 	require.Equal(t, "auto", *extractOpenAIServiceTier(map[string]any{"service_tier": "auto"}))
 	require.Equal(t, "default", *extractOpenAIServiceTier(map[string]any{"service_tier": "default"}))
 	require.Equal(t, "scale", *extractOpenAIServiceTier(map[string]any{"service_tier": "scale"}))
+	require.Equal(t, "ultrafast", *extractOpenAIServiceTier(map[string]any{"service_tier": "ultrafast"}))
 	require.Nil(t, extractOpenAIServiceTier(map[string]any{"service_tier": 1}))
 	require.Nil(t, extractOpenAIServiceTier(nil))
 }
@@ -1497,6 +1497,7 @@ func TestExtractOpenAIServiceTierFromBody(t *testing.T) {
 	require.Equal(t, "auto", *extractOpenAIServiceTierFromBody([]byte(`{"service_tier":"auto"}`)))
 	require.Equal(t, "default", *extractOpenAIServiceTierFromBody([]byte(`{"service_tier":"default"}`)))
 	require.Equal(t, "scale", *extractOpenAIServiceTierFromBody([]byte(`{"service_tier":"scale"}`)))
+	require.Equal(t, "ultrafast", *extractOpenAIServiceTierFromBody([]byte(`{"service_tier":"ultrafast"}`)))
 	require.Nil(t, extractOpenAIServiceTierFromBody([]byte(`{"service_tier":"turbo"}`)))
 	require.Nil(t, extractOpenAIServiceTierFromBody(nil))
 }

backend/internal/service/openai_gateway_request_body.go

@@ -1435,12 +1435,12 @@ func normalizeOpenAIServiceTier(raw string) *string {
 	if value == "fast" {
 		value = "priority"
 	}
-	// 放过 OpenAI 官方文档定义的所有合法 tier 值：priority/flex/auto/default/scale。
-	// 对 Codex 客户端零影响（Codex 只发 priority 或 flex，见 codex-rs/core/src/client.rs），
-	// 但能让直连 OpenAI SDK 的用户透传 auto/default/scale 以便抓包/调试。
-	// 真未知值仍返回 nil，由 normalizeResponsesBodyServiceTier 从 body 中删除。
+	// 放过 OpenAI 官方文档定义的合法 tier 值，以及 Codex/API 新增的 ultrafast。
+	// Codex 客户端会发 priority、flex 或 ultrafast；直连 OpenAI SDK 的用户还会
+	// 透传 auto/default/scale。真未知值仍返回 nil，由
+	// normalizeResponsesBodyServiceTier 从 body 中删除。
 	switch value {
-	case "priority", "flex", "auto", "default", "scale":
+	case "priority", "flex", "auto", "default", "scale", OpenAIFastTierUltrafast:
 		return &value
 	default:
 		return nil
@@ -1457,7 +1457,7 @@ type ErrInvalidOpenAIServiceTier struct {
 }
 
 func (e *ErrInvalidOpenAIServiceTier) Error() string {
-	return fmt.Sprintf("invalid service_tier %q: must be one of auto, default, fast, flex, priority, scale", e.Value)
+	return fmt.Sprintf("invalid service_tier %q: must be one of auto, default, fast, flex, priority, scale, ultrafast", e.Value)
 }
 
 const invalidOpenAIServiceTierValueMaxLen = 64
@@ -1475,7 +1475,7 @@ func boundInvalidOpenAIServiceTierValue(raw string) string {
 //   - absent / null → valid, returns "" (field omitted keeps current behavior)
 //   - "fast" → normalized to "priority" (the two are equivalent; the canonical
 //     value is what reaches the OpenAI upstream)
-//   - "priority" / "flex" / "auto" / "default" / "scale" → valid, returned as-is
+//   - "priority" / "flex" / "auto" / "default" / "scale" / "ultrafast" → valid, returned as-is
 //   - an explicitly present non-string value, an empty string, or any other
 //     unknown value → *ErrInvalidOpenAIServiceTier (handler maps to HTTP 400),
 //     matching OpenAI's enum validation semantics

backend/internal/service/openai_routing_hint.go

@@ -37,14 +37,14 @@ func setOpenAICodexRoutingHint(headers http.Header, account *Account, model stri
 
 	// Codex treats "default" as an explicit standard-routing sentinel, not as a
 	// service tier sent to the backend. Fast follows the gateway's existing
-	// canonicalization and therefore becomes "priority"; flex stays "flex".
+	// canonicalization and therefore becomes "priority"; flex and ultrafast
+	// stay as themselves.
 	canonicalTier := normalizedOpenAIServiceTierValue(serviceTier)
-	// This backport has no Codex model-catalog snapshot with which to validate
-	// arbitrary tier ids. Keep the hint to the two effective tiers Codex itself
-	// selects; default, missing, and other gateway-compatible API values remain
-	// model-only rather than expanding the ChatGPT routing protocol here.
+	// Keep the hint to the effective tiers Codex itself selects. default,
+	// missing, auto, and scale remain model-only rather than expanding the
+	// ChatGPT routing protocol here.
 	switch canonicalTier {
-	case OpenAIFastTierPriority, OpenAIFastTierFlex:
+	case OpenAIFastTierPriority, OpenAIFastTierFlex, OpenAIFastTierUltrafast:
 	default:
 		canonicalTier = ""
 	}

backend/internal/service/openai_routing_hint_test.go

@@ -24,6 +24,7 @@ func TestSetOpenAICodexRoutingHintCanonicalizesOfficialServiceTiers(t *testing.T
 		{name: "fast alias", model: "gpt-5.6", serviceTier: " fast ", want: "model=gpt-5.6;tier=priority"},
 		{name: "priority", model: "gpt-5.6", serviceTier: "priority", want: "model=gpt-5.6;tier=priority"},
 		{name: "flex", model: "gpt-5.6", serviceTier: "flex", want: "model=gpt-5.6;tier=flex"},
+		{name: "ultrafast", model: "gpt-5.6", serviceTier: "ultrafast", want: "model=gpt-5.6;tier=ultrafast"},
 		{name: "explicit default sentinel", model: "gpt-5.6", serviceTier: "default", want: "model=gpt-5.6"},
 		{name: "omitted tier", model: "gpt-5.6", want: "model=gpt-5.6"},
 		{name: "auto is not expanded without catalog support", model: "gpt-5.6", serviceTier: "auto", want: "model=gpt-5.6"},
@@ -90,6 +91,7 @@ func TestOpenAIOAuthHTTPBuildersSendRoutingHintFromFinalBody(t *testing.T) {
 	}{
 		{name: "fast", body: []byte(`{"model":"gpt-5.6-codex","service_tier":"fast"}`), want: "model=gpt-5.6-codex;tier=priority"},
 		{name: "flex", body: []byte(`{"model":"gpt-5.6-codex","service_tier":"flex"}`), want: "model=gpt-5.6-codex;tier=flex"},
+		{name: "ultrafast", body: []byte(`{"model":"gpt-5.6-codex","service_tier":"ultrafast"}`), want: "model=gpt-5.6-codex;tier=ultrafast"},
 		{name: "default", body: []byte(`{"model":"gpt-5.6-codex","service_tier":"default"}`), want: "model=gpt-5.6-codex"},
 		{name: "omitted", body: []byte(`{"model":"gpt-5.6-codex"}`), want: "model=gpt-5.6-codex"},
 	}
@@ -216,6 +218,7 @@ func TestBuildOpenAIWSHeadersSendsOAuthRoutingHintOnly(t *testing.T) {
 		},
 	}
 	require.Equal(t, "model=gpt-5.6-codex;tier=priority", build(t, oauthAccount, "fast").Get(openAICodexRoutingHintHeader))
+	require.Equal(t, "model=gpt-5.6-codex;tier=ultrafast", build(t, oauthAccount, "ultrafast").Get(openAICodexRoutingHintHeader))
 	require.Equal(t, "model=gpt-5.6-codex", build(t, oauthAccount, "default").Get(openAICodexRoutingHintHeader))
 	require.Empty(t, build(t, &Account{Platform: PlatformOpenAI, Type: AccountTypeAPIKey}, "priority").Get(openAICodexRoutingHintHeader))
 }

backend/internal/service/setting_features.go

@@ -1016,7 +1016,7 @@ func (s *SettingService) SetOpenAIFastPolicySettings(ctx context.Context, settin
 		BetaPolicyScopeAll: true, BetaPolicyScopeOAuth: true, BetaPolicyScopeAPIKey: true, BetaPolicyScopeBedrock: true,
 	}
 	validTiers := map[string]bool{
-		OpenAIFastTierAny: true, OpenAIFastTierPriority: true, OpenAIFastTierFlex: true,
+		OpenAIFastTierAny: true, OpenAIFastTierPriority: true, OpenAIFastTierUltrafast: true, OpenAIFastTierFlex: true,
 	}
 
 	for i, rule := range settings.Rules {

backend/internal/service/settings_view.go

@@ -669,15 +669,17 @@ func DefaultBetaPolicySettings() *BetaPolicySettings {
 // OpenAI Fast Policy 策略常量
 // OpenAI 的 "fast 模式" 通过请求体中的 service_tier 字段识别：
 //   - "priority"（客户端可传 "fast"，归一化为 "priority"）：fast 模式
+//   - "ultrafast"：Codex/API 的 Ultrafast 档位
 //   - "flex"：低优先级模式
 //   - 省略：normal 默认
 //
 // 本策略复用 BetaPolicyAction*/BetaPolicyScope* 常量语义，只是匹配键从
 // anthropic-beta header 换成 body 的 service_tier 字段。
 const (
-	OpenAIFastTierAny      = "all"      // 匹配任意已识别的 service_tier
-	OpenAIFastTierPriority = "priority" // 仅匹配 fast（priority）
-	OpenAIFastTierFlex     = "flex"     // 仅匹配 flex
+	OpenAIFastTierAny       = "all"       // 匹配任意已识别的 service_tier
+	OpenAIFastTierPriority  = "priority"  // 仅匹配 fast（priority）
+	OpenAIFastTierUltrafast = "ultrafast" // 仅匹配 ultrafast
+	OpenAIFastTierFlex      = "flex"      // 仅匹配 flex
 
 	// OpenAIFastPolicyActionForcePriority 会保留 service_tier 字段并强制写成
 	// priority，用于把 flex/auto/default/scale 等已识别 tier 收敛为 fast。
@@ -686,7 +688,7 @@ const (
 
 // OpenAIFastPolicyRule 单条 OpenAI fast/flex 策略规则
 type OpenAIFastPolicyRule struct {
-	ServiceTier          string   `json:"service_tier"`                     // "priority" | "flex" | "auto" | "default" | "scale" | "all"
+	ServiceTier          string   `json:"service_tier"`                     // "priority" | "ultrafast" | "flex" | "auto" | "default" | "scale" | "all"
 	Action               string   `json:"action"`                           // "pass" | "filter" | "block" | "force_priority"
 	Scope                string   `json:"scope"`                            // "all" | "oauth" | "apikey" | "bedrock"
 	UserIDs              []int64  `json:"user_ids,omitempty"`               // 空=所有 Sub2API 用户；非空=仅指定 API Key 所属用户

frontend/src/api/admin/settings.ts

@@ -1429,7 +1429,7 @@ export async function updateRectifierSettings(
  * Matches backend dto.OpenAIFastPolicyRule.
  */
 export interface OpenAIFastPolicyRule {
-  service_tier: "all" | "priority" | "flex";
+  service_tier: "all" | "priority" | "flex" | "ultrafast";
   action: "pass" | "filter" | "block" | "force_priority";
   scope: "all" | "oauth" | "apikey" | "bedrock";
   user_ids?: number[];

frontend/src/components/admin/usage/__tests__/UsageTable.spec.ts

@@ -30,6 +30,7 @@ const messages: Record<string, string> = {
   'usage.perMillionTokens': '/ 1M tokens',
   'usage.serviceTier': 'Service tier',
   'usage.serviceTierPriority': 'Fast',
+  'usage.serviceTierUltrafast': 'Ultrafast',
   'usage.serviceTierFlex': 'Flex',
   'usage.serviceTierStandard': 'Standard',
   'usage.rate': 'Rate',

frontend/src/i18n/__tests__/openaiFastPolicyLocales.spec.ts

@@ -31,6 +31,7 @@ describe('OpenAI Fast/Flex policy locale keys', () => {
   it('describes target and other-model actions without whitelist terminology', () => {
     expect(zh.admin.settings.openaiFastPolicy).toMatchObject({
       tierAll: '全部 tier 值',
+      tierUltrafast: 'ultrafast',
       modelWhitelist: '目标模型',
       fallbackAction: '其他模型处理方式',
       summaryTargetModels: '目标模型',
@@ -43,6 +44,7 @@ describe('OpenAI Fast/Flex policy locale keys', () => {
 
     expect(en.admin.settings.openaiFastPolicy).toMatchObject({
       tierAll: 'All tier values',
+      tierUltrafast: 'ultrafast',
       modelWhitelist: 'Target models',
       fallbackAction: 'Other models action',
       summaryTargetModels: 'Target models',

frontend/src/i18n/__tests__/usageServiceTierLocales.spec.ts

@@ -7,6 +7,7 @@ describe('usage service tier locale keys', () => {
   it('contains zh labels for service tier tooltip', () => {
     expect(zh.usage.serviceTier).toBe('服务档位')
     expect(zh.usage.serviceTierPriority).toBe('Fast')
+    expect(zh.usage.serviceTierUltrafast).toBe('Ultrafast')
     expect(zh.usage.serviceTierFlex).toBe('Flex')
     expect(zh.usage.serviceTierStandard).toBe('Standard')
   })
@@ -14,6 +15,7 @@ describe('usage service tier locale keys', () => {
   it('contains en labels for service tier tooltip', () => {
     expect(en.usage.serviceTier).toBe('Service tier')
     expect(en.usage.serviceTierPriority).toBe('Fast')
+    expect(en.usage.serviceTierUltrafast).toBe('Ultrafast')
     expect(en.usage.serviceTierFlex).toBe('Flex')
     expect(en.usage.serviceTierStandard).toBe('Standard')
   })

frontend/src/i18n/locales/en/admin/settings.ts

@@ -1081,7 +1081,7 @@ export default {
       },
       openaiFastPolicy: {
         title: 'OpenAI Fast/Flex Policy',
-        description: 'Intercept, filter, or pass OpenAI fast(priority) / flex requests based on the request body service_tier field. Applies to the OpenAI gateway only.',
+        description: 'Intercept, filter, or pass OpenAI fast(priority), ultrafast, or flex requests based on the request body service_tier field. Applies to the OpenAI gateway only.',
         empty: 'No rules configured. Click the button below to add one.',
         ruleHeader: 'Rule #{index}',
         removeRule: 'Remove rule',
@@ -1090,6 +1090,7 @@ export default {
         serviceTier: 'service_tier match',
         tierAll: 'All tier values',
         tierPriority: 'priority (fast)',
+        tierUltrafast: 'ultrafast',
         tierFlex: 'flex',
         action: 'Action',
         actionPass: 'Pass (keep service_tier)',

frontend/src/i18n/locales/en/dashboard.ts

@@ -398,6 +398,7 @@ export default {
     cacheWrite: 'Write',
     serviceTier: 'Service tier',
     serviceTierPriority: 'Fast',
+    serviceTierUltrafast: 'Ultrafast',
     serviceTierFlex: 'Flex',
     serviceTierStandard: 'Standard',
     rate: 'Rate',

frontend/src/i18n/locales/zh/admin/settings.ts

@@ -1075,7 +1075,7 @@ export default {
       },
       openaiFastPolicy: {
         title: 'OpenAI Fast/Flex 策略',
-        description: '基于请求体 service_tier 字段拦截/过滤/透传 OpenAI fast(priority) 与 flex 请求；仅作用于 OpenAI 网关。',
+        description: '基于请求体 service_tier 字段拦截/过滤/透传 OpenAI fast(priority)、ultrafast 与 flex 请求；仅作用于 OpenAI 网关。',
         empty: '尚未配置任何规则。点击下方按钮新增。',
         ruleHeader: '规则 #{index}',
         removeRule: '删除规则',
@@ -1084,6 +1084,7 @@ export default {
         serviceTier: 'service_tier 匹配',
         tierAll: '全部 tier 值',
         tierPriority: 'priority（fast）',
+        tierUltrafast: 'ultrafast',
         tierFlex: 'flex',
         action: '处理方式',
         actionPass: '透传（保留 service_tier）',

frontend/src/i18n/locales/zh/dashboard.ts

@@ -403,6 +403,7 @@ export default {
     cacheWrite: '写入',
     serviceTier: '服务档位',
     serviceTierPriority: 'Fast',
+    serviceTierUltrafast: 'Ultrafast',
     serviceTierFlex: 'Flex',
     serviceTierStandard: 'Standard',
     rate: '倍率',

frontend/src/utils/__tests__/usageServiceTier.spec.ts

@@ -12,6 +12,7 @@ describe('usageServiceTier utils', () => {
   it('preserves supported tiers', () => {
     expect(normalizeUsageServiceTier('priority')).toBe('priority')
     expect(normalizeUsageServiceTier('flex')).toBe('flex')
+    expect(normalizeUsageServiceTier('ultrafast')).toBe('ultrafast')
   })
 
   it('formats empty values as standard', () => {
@@ -27,11 +28,13 @@ describe('usageServiceTier utils', () => {
   it('maps tiers to translated labels', () => {
     const translate = (key: string) => ({
       'usage.serviceTierPriority': 'Fast',
+      'usage.serviceTierUltrafast': 'Ultrafast',
       'usage.serviceTierFlex': 'Flex',
       'usage.serviceTierStandard': 'Standard',
     })[key] ?? key
 
     expect(getUsageServiceTierLabel('fast', translate)).toBe('Fast')
+    expect(getUsageServiceTierLabel('ultrafast', translate)).toBe('Ultrafast')
     expect(getUsageServiceTierLabel('flex', translate)).toBe('Flex')
     expect(getUsageServiceTierLabel(undefined, translate)).toBe('Standard')
     expect(getUsageServiceTierLabel('custom-tier', translate)).toBe('custom-tier')

frontend/src/utils/usageServiceTier.ts

@@ -3,7 +3,7 @@ export function normalizeUsageServiceTier(serviceTier?: string | null): string |
   if (!value) return null
   if (value === 'fast') return 'priority'
   if (value === 'default' || value === 'standard') return 'standard'
-  if (value === 'priority' || value === 'flex') return value
+  if (value === 'priority' || value === 'flex' || value === 'ultrafast') return value
   return value
 }
 
@@ -19,6 +19,7 @@ export function getUsageServiceTierLabel(
 ): string {
   const tier = formatUsageServiceTier(serviceTier)
   if (tier === 'priority') return translate('usage.serviceTierPriority')
+  if (tier === 'ultrafast') return translate('usage.serviceTierUltrafast')
   if (tier === 'flex') return translate('usage.serviceTierFlex')
   if (tier === 'standard') return translate('usage.serviceTierStandard')
   return tier

frontend/src/views/admin/SettingsView.vue

@@ -12084,6 +12084,10 @@ const openaiFastPolicyTierOptions = computed(() => [
     value: "priority",
     label: t("admin.settings.openaiFastPolicy.tierPriority"),
   },
+  {
+    value: "ultrafast",
+    label: t("admin.settings.openaiFastPolicy.tierUltrafast"),
+  },
   { value: "flex", label: t("admin.settings.openaiFastPolicy.tierFlex") },
 ]);