From c810cad7c860b9a6af7e1dfae106c92bffa3a11c Mon Sep 17 00:00:00 2001 From: Remx Date: Thu, 19 Mar 2026 19:00:22 +0800 Subject: [PATCH 1/3] =?UTF-8?q?feat(openai):=20=E5=A2=9E=E5=8A=A0=20gpt-5.?= =?UTF-8?q?4-mini/nano=20=E6=A8=A1=E5=9E=8B=E6=94=AF=E6=8C=81=E4=B8=8E?= =?UTF-8?q?=E5=AE=9A=E4=BB=B7=E9=85=8D=E7=BD=AE?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - 接入 gpt-5.4-mini/nano 模型识别与规范化，补充默认模型列表 - 增加 gpt-5.4-mini/nano 输入/缓存命中/输出价格与计费兜底逻辑 - 同步前端模型白名单与 OpenCode 配置 - 补充 service tier(priority/flex) 计费回归测试 --- backend/internal/pkg/openai/constants.go | 2 + backend/internal/service/billing_service.go | 16 +++++ .../internal/service/billing_service_test.go | 60 +++++++++++++++++ .../service/openai_codex_transform.go | 8 +++ .../service/openai_codex_transform_test.go | 4 ++ backend/internal/service/pricing_service.go | 28 ++++++++ .../internal/service/pricing_service_test.go | 30 +++++++++ .../model_prices_and_context_window.json | 65 +++++++++++++++++++ frontend/src/components/keys/UseKeyModal.vue | 32 +++++++++ .../keys/__tests__/UseKeyModal.spec.ts | 53 +++++++++++++++ .../__tests__/useModelWhitelist.spec.ts | 11 ++++ frontend/src/composables/useModelWhitelist.ts | 2 +- 12 files changed, 310 insertions(+), 1 deletion(-) create mode 100644 frontend/src/components/keys/__tests__/UseKeyModal.spec.ts diff --git a/backend/internal/pkg/openai/constants.go b/backend/internal/pkg/openai/constants.go index b0a31a5f3..49e38bf81 100644 --- a/backend/internal/pkg/openai/constants.go +++ b/backend/internal/pkg/openai/constants.go @@ -16,6 +16,8 @@ type Model struct { // DefaultModels OpenAI models list var DefaultModels = []Model{ {ID: "gpt-5.4", Object: "model", Created: 1738368000, OwnedBy: "openai", Type: "model", DisplayName: "GPT-5.4"}, + {ID: "gpt-5.4-mini", Object: "model", Created: 1738368000, OwnedBy: "openai", Type: "model", DisplayName: "GPT-5.4 Mini"}, + {ID: "gpt-5.4-nano", Object: "model", Created: 1738368000, OwnedBy: "openai", Type: "model", DisplayName: "GPT-5.4 Nano"}, {ID: "gpt-5.3-codex", Object: "model", Created: 1735689600, OwnedBy: "openai", Type: "model", DisplayName: "GPT-5.3 Codex"}, {ID: "gpt-5.3-codex-spark", Object: "model", Created: 1735689600, OwnedBy: "openai", Type: "model", DisplayName: "GPT-5.3 Codex Spark"}, {ID: "gpt-5.2", Object: "model", Created: 1733875200, OwnedBy: "openai", Type: "model", DisplayName: "GPT-5.2"}, diff --git a/backend/internal/service/billing_service.go b/backend/internal/service/billing_service.go index 68d7a8f99..99fea0b08 100644 --- a/backend/internal/service/billing_service.go +++ b/backend/internal/service/billing_service.go @@ -221,6 +221,18 @@ func (s *BillingService) initFallbackPricing() { LongContextInputMultiplier: openAIGPT54LongContextInputMultiplier, LongContextOutputMultiplier: openAIGPT54LongContextOutputMultiplier, } + s.fallbackPrices["gpt-5.4-mini"] = &ModelPricing{ + InputPricePerToken: 7.5e-7, + OutputPricePerToken: 4.5e-6, + CacheReadPricePerToken: 7.5e-8, + SupportsCacheBreakdown: false, + } + s.fallbackPrices["gpt-5.4-nano"] = &ModelPricing{ + InputPricePerToken: 2e-7, + OutputPricePerToken: 1.25e-6, + CacheReadPricePerToken: 2e-8, + SupportsCacheBreakdown: false, + } // OpenAI GPT-5.2（本地兜底） s.fallbackPrices["gpt-5.2"] = &ModelPricing{ InputPricePerToken: 1.75e-6, @@ -294,6 +306,10 @@ func (s *BillingService) getFallbackPricing(model string) *ModelPricing { if strings.Contains(modelLower, "gpt-5") || strings.Contains(modelLower, "codex") { normalized := normalizeCodexModel(modelLower) switch normalized { + case "gpt-5.4-mini": + return s.fallbackPrices["gpt-5.4-mini"] + case "gpt-5.4-nano": + return s.fallbackPrices["gpt-5.4-nano"] case "gpt-5.4": return s.fallbackPrices["gpt-5.4"] case "gpt-5.2": diff --git a/backend/internal/service/billing_service_test.go b/backend/internal/service/billing_service_test.go index 45bbdceee..109434228 100644 --- a/backend/internal/service/billing_service_test.go +++ b/backend/internal/service/billing_service_test.go @@ -174,6 +174,30 @@ func TestGetModelPricing_OpenAIGPT54Fallback(t *testing.T) { require.InDelta(t, 1.5, pricing.LongContextOutputMultiplier, 1e-12) } +func TestGetModelPricing_OpenAIGPT54MiniFallback(t *testing.T) { + svc := newTestBillingService() + + pricing, err := svc.GetModelPricing("gpt-5.4-mini") + require.NoError(t, err) + require.NotNil(t, pricing) + require.InDelta(t, 7.5e-7, pricing.InputPricePerToken, 1e-12) + require.InDelta(t, 4.5e-6, pricing.OutputPricePerToken, 1e-12) + require.InDelta(t, 7.5e-8, pricing.CacheReadPricePerToken, 1e-12) + require.Zero(t, pricing.LongContextInputThreshold) +} + +func TestGetModelPricing_OpenAIGPT54NanoFallback(t *testing.T) { + svc := newTestBillingService() + + pricing, err := svc.GetModelPricing("gpt-5.4-nano") + require.NoError(t, err) + require.NotNil(t, pricing) + require.InDelta(t, 2e-7, pricing.InputPricePerToken, 1e-12) + require.InDelta(t, 1.25e-6, pricing.OutputPricePerToken, 1e-12) + require.InDelta(t, 2e-8, pricing.CacheReadPricePerToken, 1e-12) + require.Zero(t, pricing.LongContextInputThreshold) +} + func TestCalculateCost_OpenAIGPT54LongContextAppliesWholeSessionMultipliers(t *testing.T) { svc := newTestBillingService() @@ -210,6 +234,8 @@ func TestGetFallbackPricing_FamilyMatching(t *testing.T) { {name: "gemini unknown no fallback", model: "gemini-2.0-pro", expectNilPricing: true}, {name: "openai gpt5.1", model: "gpt-5.1", expectedInput: 1.25e-6}, {name: "openai gpt5.4", model: "gpt-5.4", expectedInput: 2.5e-6}, + {name: "openai gpt5.4 mini", model: "gpt-5.4-mini", expectedInput: 7.5e-7}, + {name: "openai gpt5.4 nano", model: "gpt-5.4-nano", expectedInput: 2e-7}, {name: "openai gpt5.3 codex", model: "gpt-5.3-codex", expectedInput: 1.5e-6}, {name: "openai gpt5.1 codex max alias", model: "gpt-5.1-codex-max", expectedInput: 1.5e-6}, {name: "openai codex mini latest alias", model: "codex-mini-latest", expectedInput: 1.5e-6}, @@ -564,6 +590,40 @@ func TestCalculateCostWithServiceTier_FlexAppliesHalfMultiplier(t *testing.T) { require.InDelta(t, baseCost.TotalCost*0.5, flexCost.TotalCost, 1e-10) } +func TestCalculateCostWithServiceTier_Gpt54MiniPriorityFallsBackToTierMultiplier(t *testing.T) { + svc := newTestBillingService() + tokens := UsageTokens{InputTokens: 120, OutputTokens: 30, CacheCreationTokens: 12, CacheReadTokens: 8} + + baseCost, err := svc.CalculateCost("gpt-5.4-mini", tokens, 1.0) + require.NoError(t, err) + + priorityCost, err := svc.CalculateCostWithServiceTier("gpt-5.4-mini", tokens, 1.0, "priority") + require.NoError(t, err) + + require.InDelta(t, baseCost.InputCost*2, priorityCost.InputCost, 1e-10) + require.InDelta(t, baseCost.OutputCost*2, priorityCost.OutputCost, 1e-10) + require.InDelta(t, baseCost.CacheCreationCost*2, priorityCost.CacheCreationCost, 1e-10) + require.InDelta(t, baseCost.CacheReadCost*2, priorityCost.CacheReadCost, 1e-10) + require.InDelta(t, baseCost.TotalCost*2, priorityCost.TotalCost, 1e-10) +} + +func TestCalculateCostWithServiceTier_Gpt54NanoFlexAppliesHalfMultiplier(t *testing.T) { + svc := newTestBillingService() + tokens := UsageTokens{InputTokens: 100, OutputTokens: 50, CacheCreationTokens: 40, CacheReadTokens: 20} + + baseCost, err := svc.CalculateCost("gpt-5.4-nano", tokens, 1.0) + require.NoError(t, err) + + flexCost, err := svc.CalculateCostWithServiceTier("gpt-5.4-nano", tokens, 1.0, "flex") + require.NoError(t, err) + + require.InDelta(t, baseCost.InputCost*0.5, flexCost.InputCost, 1e-10) + require.InDelta(t, baseCost.OutputCost*0.5, flexCost.OutputCost, 1e-10) + require.InDelta(t, baseCost.CacheCreationCost*0.5, flexCost.CacheCreationCost, 1e-10) + require.InDelta(t, baseCost.CacheReadCost*0.5, flexCost.CacheReadCost, 1e-10) + require.InDelta(t, baseCost.TotalCost*0.5, flexCost.TotalCost, 1e-10) +} + func TestCalculateCostWithServiceTier_PriorityFallsBackToTierMultiplierWithoutExplicitPriorityPrice(t *testing.T) { svc := newTestBillingService() tokens := UsageTokens{InputTokens: 120, OutputTokens: 30, CacheCreationTokens: 12, CacheReadTokens: 8} diff --git a/backend/internal/service/openai_codex_transform.go b/backend/internal/service/openai_codex_transform.go index 0ae55ad3b..d0534d8cd 100644 --- a/backend/internal/service/openai_codex_transform.go +++ b/backend/internal/service/openai_codex_transform.go @@ -7,6 +7,8 @@ import ( var codexModelMap = map[string]string{ "gpt-5.4": "gpt-5.4", + "gpt-5.4-mini": "gpt-5.4-mini", + "gpt-5.4-nano": "gpt-5.4-nano", "gpt-5.4-none": "gpt-5.4", "gpt-5.4-low": "gpt-5.4", "gpt-5.4-medium": "gpt-5.4", @@ -225,6 +227,12 @@ func normalizeCodexModel(model string) string { normalized := strings.ToLower(modelID) + if strings.Contains(normalized, "gpt-5.4-mini") || strings.Contains(normalized, "gpt 5.4 mini") { + return "gpt-5.4-mini" + } + if strings.Contains(normalized, "gpt-5.4-nano") || strings.Contains(normalized, "gpt 5.4 nano") { + return "gpt-5.4-nano" + } if strings.Contains(normalized, "gpt-5.4") || strings.Contains(normalized, "gpt 5.4") { return "gpt-5.4" } diff --git a/backend/internal/service/openai_codex_transform_test.go b/backend/internal/service/openai_codex_transform_test.go index b52f0566c..eab88c096 100644 --- a/backend/internal/service/openai_codex_transform_test.go +++ b/backend/internal/service/openai_codex_transform_test.go @@ -238,6 +238,10 @@ func TestNormalizeCodexModel_Gpt53(t *testing.T) { "gpt-5.4-high": "gpt-5.4", "gpt-5.4-chat-latest": "gpt-5.4", "gpt 5.4": "gpt-5.4", + "gpt-5.4-mini": "gpt-5.4-mini", + "gpt 5.4 mini": "gpt-5.4-mini", + "gpt-5.4-nano": "gpt-5.4-nano", + "gpt 5.4 nano": "gpt-5.4-nano", "gpt-5.3": "gpt-5.3-codex", "gpt-5.3-codex": "gpt-5.3-codex", "gpt-5.3-codex-xhigh": "gpt-5.3-codex", diff --git a/backend/internal/service/pricing_service.go b/backend/internal/service/pricing_service.go index 7ed4e7e47..10440c60a 100644 --- a/backend/internal/service/pricing_service.go +++ b/backend/internal/service/pricing_service.go @@ -34,6 +34,22 @@ var ( Mode: "chat", SupportsPromptCaching: true, } + openAIGPT54MiniFallbackPricing = &LiteLLMModelPricing{ + InputCostPerToken: 7.5e-07, + OutputCostPerToken: 4.5e-06, + CacheReadInputTokenCost: 7.5e-08, + LiteLLMProvider: "openai", + Mode: "chat", + SupportsPromptCaching: true, + } + openAIGPT54NanoFallbackPricing = &LiteLLMModelPricing{ + InputCostPerToken: 2e-07, + OutputCostPerToken: 1.25e-06, + CacheReadInputTokenCost: 2e-08, + LiteLLMProvider: "openai", + Mode: "chat", + SupportsPromptCaching: true, + } ) // LiteLLMModelPricing LiteLLM价格数据结构 @@ -723,6 +739,18 @@ func (s *PricingService) matchOpenAIModel(model string) *LiteLLMModelPricing { } } + if strings.HasPrefix(model, "gpt-5.4-mini") { + logger.With(zap.String("component", "service.pricing")). + Info(fmt.Sprintf("[Pricing] OpenAI fallback matched %s -> %s", model, "gpt-5.4-mini(static)")) + return openAIGPT54MiniFallbackPricing + } + + if strings.HasPrefix(model, "gpt-5.4-nano") { + logger.With(zap.String("component", "service.pricing")). + Info(fmt.Sprintf("[Pricing] OpenAI fallback matched %s -> %s", model, "gpt-5.4-nano(static)")) + return openAIGPT54NanoFallbackPricing + } + if strings.HasPrefix(model, "gpt-5.4") { logger.With(zap.String("component", "service.pricing")). Info(fmt.Sprintf("[Pricing] OpenAI fallback matched %s -> %s", model, "gpt-5.4(static)")) diff --git a/backend/internal/service/pricing_service_test.go b/backend/internal/service/pricing_service_test.go index 775024fdf..13a5c70c1 100644 --- a/backend/internal/service/pricing_service_test.go +++ b/backend/internal/service/pricing_service_test.go @@ -98,6 +98,36 @@ func TestGetModelPricing_Gpt54UsesStaticFallbackWhenRemoteMissing(t *testing.T) require.InDelta(t, 1.5, got.LongContextOutputCostMultiplier, 1e-12) } +func TestGetModelPricing_Gpt54MiniUsesDedicatedStaticFallbackWhenRemoteMissing(t *testing.T) { + svc := &PricingService{ + pricingData: map[string]*LiteLLMModelPricing{ + "gpt-5.1-codex": {InputCostPerToken: 1.25e-6}, + }, + } + + got := svc.GetModelPricing("gpt-5.4-mini") + require.NotNil(t, got) + require.InDelta(t, 7.5e-7, got.InputCostPerToken, 1e-12) + require.InDelta(t, 4.5e-6, got.OutputCostPerToken, 1e-12) + require.InDelta(t, 7.5e-8, got.CacheReadInputTokenCost, 1e-12) + require.Zero(t, got.LongContextInputTokenThreshold) +} + +func TestGetModelPricing_Gpt54NanoUsesDedicatedStaticFallbackWhenRemoteMissing(t *testing.T) { + svc := &PricingService{ + pricingData: map[string]*LiteLLMModelPricing{ + "gpt-5.1-codex": {InputCostPerToken: 1.25e-6}, + }, + } + + got := svc.GetModelPricing("gpt-5.4-nano") + require.NotNil(t, got) + require.InDelta(t, 2e-7, got.InputCostPerToken, 1e-12) + require.InDelta(t, 1.25e-6, got.OutputCostPerToken, 1e-12) + require.InDelta(t, 2e-8, got.CacheReadInputTokenCost, 1e-12) + require.Zero(t, got.LongContextInputTokenThreshold) +} + func TestParsePricingData_PreservesPriorityAndServiceTierFields(t *testing.T) { raw := map[string]any{ "gpt-5.4": map[string]any{ diff --git a/backend/resources/model-pricing/model_prices_and_context_window.json b/backend/resources/model-pricing/model_prices_and_context_window.json index 72860bf9e..0a096257d 100644 --- a/backend/resources/model-pricing/model_prices_and_context_window.json +++ b/backend/resources/model-pricing/model_prices_and_context_window.json @@ -5173,6 +5173,71 @@ "supports_tool_choice": true, "supports_vision": true }, + "gpt-5.4-mini": { + "cache_read_input_token_cost": 7.5e-08, + "input_cost_per_token": 7.5e-07, + "litellm_provider": "openai", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 4.5e-06, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_service_tier": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true + }, + "gpt-5.4-nano": { + "cache_read_input_token_cost": 2e-08, + "input_cost_per_token": 2e-07, + "litellm_provider": "openai", + "max_input_tokens": 400000, + "max_output_tokens": 128000, + "max_tokens": 128000, + "mode": "chat", + "output_cost_per_token": 1.25e-06, + "supported_endpoints": [ + "/v1/chat/completions", + "/v1/batch", + "/v1/responses" + ], + "supported_modalities": [ + "text", + "image" + ], + "supported_output_modalities": [ + "text" + ], + "supports_function_calling": true, + "supports_native_streaming": true, + "supports_parallel_function_calling": true, + "supports_pdf_input": true, + "supports_prompt_caching": true, + "supports_reasoning": true, + "supports_response_schema": true, + "supports_system_messages": true, + "supports_tool_choice": true, + "supports_vision": true + }, "gpt-5.3-codex": { "cache_read_input_token_cost": 1.75e-07, "cache_read_input_token_cost_priority": 3.5e-07, diff --git a/frontend/src/components/keys/UseKeyModal.vue b/frontend/src/components/keys/UseKeyModal.vue index 634db115b..7770e658b 100644 --- a/frontend/src/components/keys/UseKeyModal.vue +++ b/frontend/src/components/keys/UseKeyModal.vue @@ -709,6 +709,38 @@ function generateOpenCodeConfig(platform: string, baseUrl: string, apiKey: strin xhigh: {} } }, + 'gpt-5.4-mini': { + name: 'GPT-5.4 Mini', + limit: { + context: 400000, + output: 128000 + }, + options: { + store: false + }, + variants: { + low: {}, + medium: {}, + high: {}, + xhigh: {} + } + }, + 'gpt-5.4-nano': { + name: 'GPT-5.4 Nano', + limit: { + context: 400000, + output: 128000 + }, + options: { + store: false + }, + variants: { + low: {}, + medium: {}, + high: {}, + xhigh: {} + } + }, 'gpt-5.3-codex-spark': { name: 'GPT-5.3 Codex Spark', limit: { diff --git a/frontend/src/components/keys/__tests__/UseKeyModal.spec.ts b/frontend/src/components/keys/__tests__/UseKeyModal.spec.ts new file mode 100644 index 000000000..98b5dedec --- /dev/null +++ b/frontend/src/components/keys/__tests__/UseKeyModal.spec.ts @@ -0,0 +1,53 @@ +import { describe, expect, it, vi } from 'vitest' +import { mount } from '@vue/test-utils' +import { nextTick } from 'vue' + +vi.mock('vue-i18n', () => ({ + useI18n: () => ({ + t: (key: string) => key + }) +})) + +vi.mock('@/composables/useClipboard', () => ({ + useClipboard: () => ({ + copyToClipboard: vi.fn().mockResolvedValue(true) + }) +})) + +import UseKeyModal from '../UseKeyModal.vue' + +describe('UseKeyModal', () => { + it('renders updated GPT-5.4 mini/nano names in OpenCode config', async () => { + const wrapper = mount(UseKeyModal, { + props: { + show: true, + apiKey: 'sk-test', + baseUrl: 'https://example.com/v1', + platform: 'openai' + }, + global: { + stubs: { + BaseDialog: { + template: '
' + }, + Icon: { + template: '' + } + } + } + }) + + const opencodeTab = wrapper.findAll('button').find((button) => + button.text().includes('keys.useKeyModal.cliTabs.opencode') + ) + + expect(opencodeTab).toBeDefined() + await opencodeTab!.trigger('click') + await nextTick() + + const codeBlock = wrapper.find('pre code') + expect(codeBlock.exists()).toBe(true) + expect(codeBlock.text()).toContain('"name": "GPT-5.4 Mini"') + expect(codeBlock.text()).toContain('"name": "GPT-5.4 Nano"') + }) +}) diff --git a/frontend/src/composables/__tests__/useModelWhitelist.spec.ts b/frontend/src/composables/__tests__/useModelWhitelist.spec.ts index b4308a634..4061be4dd 100644 --- a/frontend/src/composables/__tests__/useModelWhitelist.spec.ts +++ b/frontend/src/composables/__tests__/useModelWhitelist.spec.ts @@ -11,6 +11,8 @@ describe('useModelWhitelist', () => { const models = getModelsByPlatform('openai') expect(models).toContain('gpt-5.4') + expect(models).toContain('gpt-5.4-mini') + expect(models).toContain('gpt-5.4-nano') expect(models).toContain('gpt-5.4-2026-03-05') }) @@ -52,4 +54,13 @@ describe('useModelWhitelist', () => { 'gpt-5.4-2026-03-05': 'gpt-5.4-2026-03-05' }) }) + + it('whitelist keeps GPT-5.4 mini and nano exact mappings', () => { + const mapping = buildModelMappingObject('whitelist', ['gpt-5.4-mini', 'gpt-5.4-nano'], []) + + expect(mapping).toEqual({ + 'gpt-5.4-mini': 'gpt-5.4-mini', + 'gpt-5.4-nano': 'gpt-5.4-nano' + }) + }) }) diff --git a/frontend/src/composables/useModelWhitelist.ts b/frontend/src/composables/useModelWhitelist.ts index 0ff288bb7..9e7cb036d 100644 --- a/frontend/src/composables/useModelWhitelist.ts +++ b/frontend/src/composables/useModelWhitelist.ts @@ -25,7 +25,7 @@ const openaiModels = [ 'gpt-5.2', 'gpt-5.2-2025-12-11', 'gpt-5.2-chat-latest', 'gpt-5.2-codex', 'gpt-5.2-pro', 'gpt-5.2-pro-2025-12-11', // GPT-5.4 系列 - 'gpt-5.4', 'gpt-5.4-2026-03-05', + 'gpt-5.4', 'gpt-5.4-mini', 'gpt-5.4-nano', 'gpt-5.4-2026-03-05', // GPT-5.3 系列 'gpt-5.3-codex', 'gpt-5.3-codex-spark', 'chatgpt-4o-latest', From 995ef1348a63d6ee238cbb5fd9aa4ff4cb1d6fcc Mon Sep 17 00:00:00 2001 From: InCerry Date: Tue, 24 Mar 2026 19:20:15 +0800 Subject: [PATCH 2/3] refactor: improve model resolution and normalization logic for OpenAI integration --- .../service/openai_codex_transform.go | 2 +- .../service/openai_codex_transform_test.go | 29 ++++++++++++++ .../service/openai_compat_prompt_cache_key.go | 8 ++-- .../openai_compat_prompt_cache_key_test.go | 15 +++++++ .../openai_gateway_chat_completions.go | 33 +++++++++------ .../service/openai_gateway_messages.go | 29 +++++++++----- .../service/openai_gateway_service.go | 40 +++++++++---------- .../internal/service/openai_model_mapping.go | 28 +++++++++++-- .../service/openai_model_mapping_test.go | 29 +++++++++++--- .../internal/service/openai_ws_forwarder.go | 14 ++----- 10 files changed, 159 insertions(+), 68 deletions(-) diff --git a/backend/internal/service/openai_codex_transform.go b/backend/internal/service/openai_codex_transform.go index d0534d8cd..21b4874eb 100644 --- a/backend/internal/service/openai_codex_transform.go +++ b/backend/internal/service/openai_codex_transform.go @@ -85,7 +85,7 @@ func applyCodexOAuthTransform(reqBody map[string]any, isCodexCLI bool, isCompact if v, ok := reqBody["model"].(string); ok { model = v } - normalizedModel := normalizeCodexModel(model) + normalizedModel := strings.TrimSpace(model) if normalizedModel != "" { if model != normalizedModel { reqBody["model"] = normalizedModel diff --git a/backend/internal/service/openai_codex_transform_test.go b/backend/internal/service/openai_codex_transform_test.go index eab88c096..889ac6159 100644 --- a/backend/internal/service/openai_codex_transform_test.go +++ b/backend/internal/service/openai_codex_transform_test.go @@ -246,6 +246,7 @@ func TestNormalizeCodexModel_Gpt53(t *testing.T) { "gpt-5.3-codex": "gpt-5.3-codex", "gpt-5.3-codex-xhigh": "gpt-5.3-codex", "gpt-5.3-codex-spark": "gpt-5.3-codex", + "gpt 5.3 codex spark": "gpt-5.3-codex", "gpt-5.3-codex-spark-high": "gpt-5.3-codex", "gpt-5.3-codex-spark-xhigh": "gpt-5.3-codex", "gpt 5.3 codex": "gpt-5.3-codex", @@ -256,6 +257,34 @@ func TestNormalizeCodexModel_Gpt53(t *testing.T) { } } +func TestApplyCodexOAuthTransform_PreservesBareSparkModel(t *testing.T) { + reqBody := map[string]any{ + "model": "gpt-5.3-codex-spark", + "input": []any{}, + } + + result := applyCodexOAuthTransform(reqBody, false, false) + + require.Equal(t, "gpt-5.3-codex-spark", reqBody["model"]) + require.Equal(t, "gpt-5.3-codex-spark", result.NormalizedModel) + store, ok := reqBody["store"].(bool) + require.True(t, ok) + require.False(t, store) +} + +func TestApplyCodexOAuthTransform_TrimmedModelWithoutPolicyRewrite(t *testing.T) { + reqBody := map[string]any{ + "model": " gpt-5.3-codex-spark ", + "input": []any{}, + } + + result := applyCodexOAuthTransform(reqBody, false, false) + + require.Equal(t, "gpt-5.3-codex-spark", reqBody["model"]) + require.Equal(t, "gpt-5.3-codex-spark", result.NormalizedModel) + require.True(t, result.Modified) +} + func TestApplyCodexOAuthTransform_CodexCLI_PreservesExistingInstructions(t *testing.T) { // Codex CLI 场景：已有 instructions 时不修改 diff --git a/backend/internal/service/openai_compat_prompt_cache_key.go b/backend/internal/service/openai_compat_prompt_cache_key.go index 88e16a4db..46381838a 100644 --- a/backend/internal/service/openai_compat_prompt_cache_key.go +++ b/backend/internal/service/openai_compat_prompt_cache_key.go @@ -10,8 +10,8 @@ import ( const compatPromptCacheKeyPrefix = "compat_cc_" func shouldAutoInjectPromptCacheKeyForCompat(model string) bool { - switch normalizeCodexModel(strings.TrimSpace(model)) { - case "gpt-5.4", "gpt-5.3-codex": + switch resolveOpenAIUpstreamModel(strings.TrimSpace(model)) { + case "gpt-5.4", "gpt-5.3-codex", "gpt-5.3-codex-spark": return true default: return false @@ -23,9 +23,9 @@ func deriveCompatPromptCacheKey(req *apicompat.ChatCompletionsRequest, mappedMod return "" } - normalizedModel := normalizeCodexModel(strings.TrimSpace(mappedModel)) + normalizedModel := resolveOpenAIUpstreamModel(strings.TrimSpace(mappedModel)) if normalizedModel == "" { - normalizedModel = normalizeCodexModel(strings.TrimSpace(req.Model)) + normalizedModel = resolveOpenAIUpstreamModel(strings.TrimSpace(req.Model)) } if normalizedModel == "" { normalizedModel = strings.TrimSpace(req.Model) diff --git a/backend/internal/service/openai_compat_prompt_cache_key_test.go b/backend/internal/service/openai_compat_prompt_cache_key_test.go index eb9148de2..6ca3e85cd 100644 --- a/backend/internal/service/openai_compat_prompt_cache_key_test.go +++ b/backend/internal/service/openai_compat_prompt_cache_key_test.go @@ -17,6 +17,7 @@ func TestShouldAutoInjectPromptCacheKeyForCompat(t *testing.T) { require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.4")) require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.3")) require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.3-codex")) + require.True(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-5.3-codex-spark")) require.False(t, shouldAutoInjectPromptCacheKeyForCompat("gpt-4o")) } @@ -62,3 +63,17 @@ func TestDeriveCompatPromptCacheKey_DiffersAcrossSessions(t *testing.T) { k2 := deriveCompatPromptCacheKey(req2, "gpt-5.4") require.NotEqual(t, k1, k2, "different first user messages should yield different keys") } + +func TestDeriveCompatPromptCacheKey_UsesResolvedSparkFamily(t *testing.T) { + req := &apicompat.ChatCompletionsRequest{ + Model: "gpt-5.3-codex-spark", + Messages: []apicompat.ChatMessage{ + {Role: "user", Content: mustRawJSON(t, `"Question A"`)}, + }, + } + + k1 := deriveCompatPromptCacheKey(req, "gpt-5.3-codex-spark") + k2 := deriveCompatPromptCacheKey(req, " openai/gpt-5.3-codex-spark ") + require.NotEmpty(t, k1) + require.Equal(t, k1, k2, "resolved spark family should derive a stable compat cache key") +} diff --git a/backend/internal/service/openai_gateway_chat_completions.go b/backend/internal/service/openai_gateway_chat_completions.go index a442da33b..1d5bf0d0a 100644 --- a/backend/internal/service/openai_gateway_chat_completions.go +++ b/backend/internal/service/openai_gateway_chat_completions.go @@ -45,12 +45,13 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions( // 2. Resolve model mapping early so compat prompt_cache_key injection can // derive a stable seed from the final upstream model family. - mappedModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel) + billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel) + upstreamModel := resolveOpenAIUpstreamModel(billingModel) promptCacheKey = strings.TrimSpace(promptCacheKey) compatPromptCacheInjected := false - if promptCacheKey == "" && account.Type == AccountTypeOAuth && shouldAutoInjectPromptCacheKeyForCompat(mappedModel) { - promptCacheKey = deriveCompatPromptCacheKey(&chatReq, mappedModel) + if promptCacheKey == "" && account.Type == AccountTypeOAuth && shouldAutoInjectPromptCacheKeyForCompat(upstreamModel) { + promptCacheKey = deriveCompatPromptCacheKey(&chatReq, upstreamModel) compatPromptCacheInjected = promptCacheKey != "" } @@ -60,12 +61,13 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions( if err != nil { return nil, fmt.Errorf("convert chat completions to responses: %w", err) } - responsesReq.Model = mappedModel + responsesReq.Model = upstreamModel logFields := []zap.Field{ zap.Int64("account_id", account.ID), zap.String("original_model", originalModel), - zap.String("mapped_model", mappedModel), + zap.String("billing_model", billingModel), + zap.String("upstream_model", upstreamModel), zap.Bool("stream", clientStream), } if compatPromptCacheInjected { @@ -88,6 +90,9 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions( return nil, fmt.Errorf("unmarshal for codex transform: %w", err) } codexResult := applyCodexOAuthTransform(reqBody, false, false) + if codexResult.NormalizedModel != "" { + upstreamModel = codexResult.NormalizedModel + } if codexResult.PromptCacheKey != "" { promptCacheKey = codexResult.PromptCacheKey } else if promptCacheKey != "" { @@ -180,9 +185,9 @@ func (s *OpenAIGatewayService) ForwardAsChatCompletions( var result *OpenAIForwardResult var handleErr error if clientStream { - result, handleErr = s.handleChatStreamingResponse(resp, c, originalModel, mappedModel, includeUsage, startTime) + result, handleErr = s.handleChatStreamingResponse(resp, c, originalModel, billingModel, upstreamModel, includeUsage, startTime) } else { - result, handleErr = s.handleChatBufferedStreamingResponse(resp, c, originalModel, mappedModel, startTime) + result, handleErr = s.handleChatBufferedStreamingResponse(resp, c, originalModel, billingModel, upstreamModel, startTime) } // Propagate ServiceTier and ReasoningEffort to result for billing @@ -224,7 +229,8 @@ func (s *OpenAIGatewayService) handleChatBufferedStreamingResponse( resp *http.Response, c *gin.Context, originalModel string, - mappedModel string, + billingModel string, + upstreamModel string, startTime time.Time, ) (*OpenAIForwardResult, error) { requestID := resp.Header.Get("x-request-id") @@ -295,8 +301,8 @@ func (s *OpenAIGatewayService) handleChatBufferedStreamingResponse( RequestID: requestID, Usage: usage, Model: originalModel, - BillingModel: mappedModel, - UpstreamModel: mappedModel, + BillingModel: billingModel, + UpstreamModel: upstreamModel, Stream: false, Duration: time.Since(startTime), }, nil @@ -308,7 +314,8 @@ func (s *OpenAIGatewayService) handleChatStreamingResponse( resp *http.Response, c *gin.Context, originalModel string, - mappedModel string, + billingModel string, + upstreamModel string, includeUsage bool, startTime time.Time, ) (*OpenAIForwardResult, error) { @@ -343,8 +350,8 @@ func (s *OpenAIGatewayService) handleChatStreamingResponse( RequestID: requestID, Usage: usage, Model: originalModel, - BillingModel: mappedModel, - UpstreamModel: mappedModel, + BillingModel: billingModel, + UpstreamModel: upstreamModel, Stream: true, Duration: time.Since(startTime), FirstTokenMs: firstTokenMs, diff --git a/backend/internal/service/openai_gateway_messages.go b/backend/internal/service/openai_gateway_messages.go index 6a29823ae..e9548b79a 100644 --- a/backend/internal/service/openai_gateway_messages.go +++ b/backend/internal/service/openai_gateway_messages.go @@ -59,13 +59,15 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic( } // 3. Model mapping - mappedModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel) - responsesReq.Model = mappedModel + billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel) + upstreamModel := resolveOpenAIUpstreamModel(billingModel) + responsesReq.Model = upstreamModel logger.L().Debug("openai messages: model mapping applied", zap.Int64("account_id", account.ID), zap.String("original_model", originalModel), - zap.String("mapped_model", mappedModel), + zap.String("billing_model", billingModel), + zap.String("upstream_model", upstreamModel), zap.Bool("stream", isStream), ) @@ -81,6 +83,9 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic( return nil, fmt.Errorf("unmarshal for codex transform: %w", err) } codexResult := applyCodexOAuthTransform(reqBody, false, false) + if codexResult.NormalizedModel != "" { + upstreamModel = codexResult.NormalizedModel + } if codexResult.PromptCacheKey != "" { promptCacheKey = codexResult.PromptCacheKey } else if promptCacheKey != "" { @@ -181,10 +186,10 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic( var result *OpenAIForwardResult var handleErr error if clientStream { - result, handleErr = s.handleAnthropicStreamingResponse(resp, c, originalModel, mappedModel, startTime) + result, handleErr = s.handleAnthropicStreamingResponse(resp, c, originalModel, billingModel, upstreamModel, startTime) } else { // Client wants JSON: buffer the streaming response and assemble a JSON reply. - result, handleErr = s.handleAnthropicBufferedStreamingResponse(resp, c, originalModel, mappedModel, startTime) + result, handleErr = s.handleAnthropicBufferedStreamingResponse(resp, c, originalModel, billingModel, upstreamModel, startTime) } // Propagate ServiceTier and ReasoningEffort to result for billing @@ -229,7 +234,8 @@ func (s *OpenAIGatewayService) handleAnthropicBufferedStreamingResponse( resp *http.Response, c *gin.Context, originalModel string, - mappedModel string, + billingModel string, + upstreamModel string, startTime time.Time, ) (*OpenAIForwardResult, error) { requestID := resp.Header.Get("x-request-id") @@ -302,8 +308,8 @@ func (s *OpenAIGatewayService) handleAnthropicBufferedStreamingResponse( RequestID: requestID, Usage: usage, Model: originalModel, - BillingModel: mappedModel, - UpstreamModel: mappedModel, + BillingModel: billingModel, + UpstreamModel: upstreamModel, Stream: false, Duration: time.Since(startTime), }, nil @@ -318,7 +324,8 @@ func (s *OpenAIGatewayService) handleAnthropicStreamingResponse( resp *http.Response, c *gin.Context, originalModel string, - mappedModel string, + billingModel string, + upstreamModel string, startTime time.Time, ) (*OpenAIForwardResult, error) { requestID := resp.Header.Get("x-request-id") @@ -351,8 +358,8 @@ func (s *OpenAIGatewayService) handleAnthropicStreamingResponse( RequestID: requestID, Usage: usage, Model: originalModel, - BillingModel: mappedModel, - UpstreamModel: mappedModel, + BillingModel: billingModel, + UpstreamModel: upstreamModel, Stream: true, Duration: time.Since(startTime), FirstTokenMs: firstTokenMs, diff --git a/backend/internal/service/openai_gateway_service.go b/backend/internal/service/openai_gateway_service.go index 4e96cf059..daccf38ff 100644 --- a/backend/internal/service/openai_gateway_service.go +++ b/backend/internal/service/openai_gateway_service.go @@ -1778,29 +1778,29 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco } // 对所有请求执行模型映射（包含 Codex CLI）。 - mappedModel := account.GetMappedModel(reqModel) - if mappedModel != reqModel { - logger.LegacyPrintf("service.openai_gateway", "[OpenAI] Model mapping applied: %s -> %s (account: %s, isCodexCLI: %v)", reqModel, mappedModel, account.Name, isCodexCLI) - reqBody["model"] = mappedModel + billingModel := account.GetMappedModel(reqModel) + if billingModel != reqModel { + logger.LegacyPrintf("service.openai_gateway", "[OpenAI] Model mapping applied: %s -> %s (account: %s, isCodexCLI: %v)", reqModel, billingModel, account.Name, isCodexCLI) + reqBody["model"] = billingModel bodyModified = true - markPatchSet("model", mappedModel) + markPatchSet("model", billingModel) } + upstreamModel := billingModel // 针对所有 OpenAI 账号执行 Codex 模型名规范化，确保上游识别一致。 if model, ok := reqBody["model"].(string); ok { - normalizedModel := normalizeCodexModel(model) - if normalizedModel != "" && normalizedModel != model { - logger.LegacyPrintf("service.openai_gateway", "[OpenAI] Codex model normalization: %s -> %s (account: %s, type: %s, isCodexCLI: %v)", - model, normalizedModel, account.Name, account.Type, isCodexCLI) - reqBody["model"] = normalizedModel - mappedModel = normalizedModel + upstreamModel = resolveOpenAIUpstreamModel(model) + if upstreamModel != "" && upstreamModel != model { + logger.LegacyPrintf("service.openai_gateway", "[OpenAI] Upstream model resolved: %s -> %s (account: %s, type: %s, isCodexCLI: %v)", + model, upstreamModel, account.Name, account.Type, isCodexCLI) + reqBody["model"] = upstreamModel bodyModified = true - markPatchSet("model", normalizedModel) + markPatchSet("model", upstreamModel) } // 移除 gpt-5.2-codex 以下的版本 verbosity 参数 // 确保高版本模型向低版本模型映射不报错 - if !SupportsVerbosity(normalizedModel) { + if !SupportsVerbosity(upstreamModel) { if text, ok := reqBody["text"].(map[string]any); ok { delete(text, "verbosity") } @@ -1824,7 +1824,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco disablePatch() } if codexResult.NormalizedModel != "" { - mappedModel = codexResult.NormalizedModel + upstreamModel = codexResult.NormalizedModel } if codexResult.PromptCacheKey != "" { promptCacheKey = codexResult.PromptCacheKey @@ -1941,7 +1941,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco "forward_start account_id=%d account_type=%s model=%s stream=%v has_previous_response_id=%v", account.ID, account.Type, - mappedModel, + upstreamModel, reqStream, hasPreviousResponseID, ) @@ -2030,7 +2030,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco isCodexCLI, reqStream, originalModel, - mappedModel, + upstreamModel, startTime, attempt, wsLastFailureReason, @@ -2131,7 +2131,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco firstTokenMs, wsAttempts, ) - wsResult.UpstreamModel = mappedModel + wsResult.UpstreamModel = upstreamModel return wsResult, nil } s.writeOpenAIWSFallbackErrorResponse(c, account, wsErr) @@ -2236,14 +2236,14 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco var usage *OpenAIUsage var firstTokenMs *int if reqStream { - streamResult, err := s.handleStreamingResponse(ctx, resp, c, account, startTime, originalModel, mappedModel) + streamResult, err := s.handleStreamingResponse(ctx, resp, c, account, startTime, originalModel, upstreamModel) if err != nil { return nil, err } usage = streamResult.usage firstTokenMs = streamResult.firstTokenMs } else { - usage, err = s.handleNonStreamingResponse(ctx, resp, c, account, originalModel, mappedModel) + usage, err = s.handleNonStreamingResponse(ctx, resp, c, account, originalModel, upstreamModel) if err != nil { return nil, err } @@ -2267,7 +2267,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco RequestID: resp.Header.Get("x-request-id"), Usage: *usage, Model: originalModel, - UpstreamModel: mappedModel, + UpstreamModel: upstreamModel, ServiceTier: serviceTier, ReasoningEffort: reasoningEffort, Stream: reqStream, diff --git a/backend/internal/service/openai_model_mapping.go b/backend/internal/service/openai_model_mapping.go index 9bf3fba3b..4f8c094bc 100644 --- a/backend/internal/service/openai_model_mapping.go +++ b/backend/internal/service/openai_model_mapping.go @@ -1,8 +1,10 @@ package service -// resolveOpenAIForwardModel determines the upstream model for OpenAI-compatible -// forwarding. Group-level default mapping only applies when the account itself -// did not match any explicit model_mapping rule. +import "strings" + +// resolveOpenAIForwardModel resolves the account/group mapping result for +// OpenAI-compatible forwarding. Group-level default mapping only applies when +// the account itself did not match any explicit model_mapping rule. func resolveOpenAIForwardModel(account *Account, requestedModel, defaultMappedModel string) string { if account == nil { if defaultMappedModel != "" { @@ -17,3 +19,23 @@ func resolveOpenAIForwardModel(account *Account, requestedModel, defaultMappedMo } return mappedModel } + +func resolveOpenAIUpstreamModel(model string) string { + if isBareGPT53CodexSparkModel(model) { + return "gpt-5.3-codex-spark" + } + return normalizeCodexModel(strings.TrimSpace(model)) +} + +func isBareGPT53CodexSparkModel(model string) bool { + modelID := strings.TrimSpace(model) + if modelID == "" { + return false + } + if strings.Contains(modelID, "/") { + parts := strings.Split(modelID, "/") + modelID = parts[len(parts)-1] + } + normalized := strings.ToLower(strings.TrimSpace(modelID)) + return normalized == "gpt-5.3-codex-spark" || normalized == "gpt 5.3 codex spark" +} diff --git a/backend/internal/service/openai_model_mapping_test.go b/backend/internal/service/openai_model_mapping_test.go index edbb968bd..42f58b374 100644 --- a/backend/internal/service/openai_model_mapping_test.go +++ b/backend/internal/service/openai_model_mapping_test.go @@ -74,13 +74,30 @@ func TestResolveOpenAIForwardModel_PreventsClaudeModelFromFallingBackToGpt51(t * Credentials: map[string]any{}, } - withoutDefault := resolveOpenAIForwardModel(account, "claude-opus-4-6", "") - if got := normalizeCodexModel(withoutDefault); got != "gpt-5.1" { - t.Fatalf("normalizeCodexModel(%q) = %q, want %q", withoutDefault, got, "gpt-5.1") + withoutDefault := resolveOpenAIUpstreamModel(resolveOpenAIForwardModel(account, "claude-opus-4-6", "")) + if withoutDefault != "gpt-5.1" { + t.Fatalf("resolveOpenAIUpstreamModel(...) = %q, want %q", withoutDefault, "gpt-5.1") } - withDefault := resolveOpenAIForwardModel(account, "claude-opus-4-6", "gpt-5.4") - if got := normalizeCodexModel(withDefault); got != "gpt-5.4" { - t.Fatalf("normalizeCodexModel(%q) = %q, want %q", withDefault, got, "gpt-5.4") + withDefault := resolveOpenAIUpstreamModel(resolveOpenAIForwardModel(account, "claude-opus-4-6", "gpt-5.4")) + if withDefault != "gpt-5.4" { + t.Fatalf("resolveOpenAIUpstreamModel(...) = %q, want %q", withDefault, "gpt-5.4") + } +} + +func TestResolveOpenAIUpstreamModel(t *testing.T) { + cases := map[string]string{ + "gpt-5.3-codex-spark": "gpt-5.3-codex-spark", + "gpt 5.3 codex spark": "gpt-5.3-codex-spark", + " openai/gpt-5.3-codex-spark ": "gpt-5.3-codex-spark", + "gpt-5.3-codex-spark-high": "gpt-5.3-codex", + "gpt-5.3-codex-spark-xhigh": "gpt-5.3-codex", + "gpt-5.3": "gpt-5.3-codex", + } + + for input, expected := range cases { + if got := resolveOpenAIUpstreamModel(input); got != expected { + t.Fatalf("resolveOpenAIUpstreamModel(%q) = %q, want %q", input, got, expected) + } } } diff --git a/backend/internal/service/openai_ws_forwarder.go b/backend/internal/service/openai_ws_forwarder.go index 814ec0bde..9c30a3906 100644 --- a/backend/internal/service/openai_ws_forwarder.go +++ b/backend/internal/service/openai_ws_forwarder.go @@ -2515,12 +2515,9 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient( } normalized = next } - mappedModel := account.GetMappedModel(originalModel) - if normalizedModel := normalizeCodexModel(mappedModel); normalizedModel != "" { - mappedModel = normalizedModel - } - if mappedModel != originalModel { - next, setErr := applyPayloadMutation(normalized, "model", mappedModel) + upstreamModel := resolveOpenAIUpstreamModel(account.GetMappedModel(originalModel)) + if upstreamModel != originalModel { + next, setErr := applyPayloadMutation(normalized, "model", upstreamModel) if setErr != nil { return openAIWSClientPayload{}, NewOpenAIWSClientCloseError(coderws.StatusPolicyViolation, "invalid websocket request payload", setErr) } @@ -2776,10 +2773,7 @@ func (s *OpenAIGatewayService) ProxyResponsesWebSocketFromClient( mappedModel := "" var mappedModelBytes []byte if originalModel != "" { - mappedModel = account.GetMappedModel(originalModel) - if normalizedModel := normalizeCodexModel(mappedModel); normalizedModel != "" { - mappedModel = normalizedModel - } + mappedModel = resolveOpenAIUpstreamModel(account.GetMappedModel(originalModel)) needModelReplace = mappedModel != "" && mappedModel != originalModel if needModelReplace { mappedModelBytes = []byte(mappedModel) From 0b3feb9d4c85df45d83b3165cc35de8cda67d6b3 Mon Sep 17 00:00:00 2001 From: InCerry Date: Tue, 31 Mar 2026 10:33:28 +0800 Subject: [PATCH 3/3] fix(openai): resolve Anthropic compat mapping from normalized model Anthropic compat requests normalize reasoning suffixes before forwarding, but the account mapping step was still using the raw request model. Resolve billing and upstream models from the normalized compat model so explicit account mappings win over fallback defaults. Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent) Co-authored-by: Sisyphus --- backend/internal/service/openai_gateway_messages.go | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/backend/internal/service/openai_gateway_messages.go b/backend/internal/service/openai_gateway_messages.go index 02efc23b1..8c389556f 100644 --- a/backend/internal/service/openai_gateway_messages.go +++ b/backend/internal/service/openai_gateway_messages.go @@ -41,6 +41,7 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic( } originalModel := anthropicReq.Model applyOpenAICompatModelNormalization(&anthropicReq) + normalizedModel := anthropicReq.Model clientStream := anthropicReq.Stream // client's original stream preference // 2. Convert Anthropic → Responses @@ -60,13 +61,14 @@ func (s *OpenAIGatewayService) ForwardAsAnthropic( } // 3. Model mapping - billingModel := resolveOpenAIForwardModel(account, originalModel, defaultMappedModel) + billingModel := resolveOpenAIForwardModel(account, normalizedModel, defaultMappedModel) upstreamModel := resolveOpenAIUpstreamModel(billingModel) responsesReq.Model = upstreamModel logger.L().Debug("openai messages: model mapping applied", zap.Int64("account_id", account.ID), zap.String("original_model", originalModel), + zap.String("normalized_model", normalizedModel), zap.String("billing_model", billingModel), zap.String("upstream_model", upstreamModel), zap.Bool("stream", isStream),