🔨 chore: New API support switch Responses API mode

packages/model-runtime/src/core/openaiCompatibleFactory/index.ts

@@ -208,32 +208,39 @@ export const createOpenAICompatibleRuntime = <T extends Record<string, any> = an
 
         log('chat called with model: %s, stream: %s', payload.model, payload.stream ?? true);
 
-        // 工厂级 Responses API 路由控制（支持实例覆盖）
-        const modelId = (payload as any).model as string | undefined;
-        const shouldUseResponses = (() => {
+        let processedPayload: any = payload;
+        const userApiMode = (payload as any).apiMode;
+
+        if (userApiMode === 'responses') {
+          // User switch is ON, check if model is in the whitelist
+          const modelId = (payload as any).model as string | undefined;
           const instanceChat = ((this._options as any).chatCompletion || {}) as {
             useResponse?: boolean;
             useResponseModels?: Array<string | RegExp>;
           };
-          const flagUseResponse =
-            instanceChat.useResponse ?? (chatCompletion ? chatCompletion.useResponse : undefined);
           const flagUseResponseModels =
             instanceChat.useResponseModels ?? chatCompletion?.useResponseModels;
 
-          if (!chatCompletion && !instanceChat) return false;
-          if (flagUseResponse) return true;
-          if (!modelId || !flagUseResponseModels?.length) return false;
-          return flagUseResponseModels.some((m: string | RegExp) =>
-            typeof m === 'string' ? modelId.includes(m) : (m as RegExp).test(modelId),
-          );
-        })();
-
-        let processedPayload: any = payload;
-        if (shouldUseResponses) {
-          log('using Responses API mode');
-          processedPayload = { ...payload, apiMode: 'responses' } as any;
+          // Check if model matches useResponseModels whitelist
+          const modelInWhitelist = (() => {
+            if (!modelId) return false;
+            if (!flagUseResponseModels?.length) return true; // No whitelist = allow all
+            return flagUseResponseModels.some((m: string | RegExp) =>
+              typeof m === 'string' ? modelId.includes(m) : (m as RegExp).test(modelId),
+            );
+          })();
+
+          if (modelInWhitelist) {
+            log('using Responses API mode (switch ON + model in whitelist)');
+            // Keep apiMode: 'responses'
+          } else {
+            log('using Chat Completions API mode (switch ON but model not in whitelist)');
+            processedPayload = { ...payload, apiMode: undefined } as any;
+          }
         } else {
-          log('using Chat Completions API mode');
+          // User switch is OFF or not set, always use Chat Completions API
+          log('using Chat Completions API mode (switch OFF)');
+          processedPayload = { ...payload, apiMode: undefined } as any;
         }
 
         // 再进行工厂级处理

packages/model-runtime/src/providers/newapi/index.test.ts

@@ -5,7 +5,7 @@ import { Mock, afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
 import { responsesAPIModels } from '../../const/models';
 import { ChatStreamPayload } from '../../types/chat';
 import * as modelParseModule from '../../utils/modelParse';
-import { LobeNewAPIAI, NewAPIModelCard, NewAPIPricing, handlePayload, params } from './index';
+import { LobeNewAPIAI, NewAPIModelCard, NewAPIPricing, params } from './index';
 
 // Mock external dependencies
 vi.mock('../../utils/modelParse');
@@ -701,78 +701,6 @@ describe('NewAPI Runtime - 100% Branch Coverage', () => {
     });
   });
 
-  describe('HandlePayload Function - Direct Testing', () => {
-    beforeEach(() => {
-      // Mock responsesAPIModels as a Set for testing
-      (responsesAPIModels as any).has = vi.fn((model: string) => model === 'o1-pro');
-    });
-
-    it('should add apiMode for models in responsesAPIModels set', () => {
-      (responsesAPIModels as any).has = vi.fn((model: string) => model === 'o1-pro');
-
-      const payload: ChatStreamPayload = {
-        model: 'o1-pro',
-        messages: [{ role: 'user', content: 'test' }],
-        temperature: 0.5,
-      };
-
-      const result = handlePayload(payload);
-      expect(result).toEqual({ ...payload, apiMode: 'responses' });
-    });
-
-    it('should add apiMode for gpt- models', () => {
-      (responsesAPIModels as any).has = vi.fn(() => false);
-
-      const payload: ChatStreamPayload = {
-        model: 'gpt-4o',
-        messages: [{ role: 'user', content: 'test' }],
-        temperature: 0.5,
-      };
-
-      const result = handlePayload(payload);
-      expect(result).toEqual({ ...payload, apiMode: 'responses' });
-    });
-
-    it('should add apiMode for o1 models', () => {
-      (responsesAPIModels as any).has = vi.fn(() => false);
-
-      const payload: ChatStreamPayload = {
-        model: 'o1-mini',
-        messages: [{ role: 'user', content: 'test' }],
-        temperature: 0.5,
-      };
-
-      const result = handlePayload(payload);
-      expect(result).toEqual({ ...payload, apiMode: 'responses' });
-    });
-
-    it('should add apiMode for o3 models', () => {
-      (responsesAPIModels as any).has = vi.fn(() => false);
-
-      const payload: ChatStreamPayload = {
-        model: 'o3-turbo',
-        messages: [{ role: 'user', content: 'test' }],
-        temperature: 0.5,
-      };
-
-      const result = handlePayload(payload);
-      expect(result).toEqual({ ...payload, apiMode: 'responses' });
-    });
-
-    it('should not modify payload for regular models', () => {
-      (responsesAPIModels as any).has = vi.fn(() => false);
-
-      const payload: ChatStreamPayload = {
-        model: 'claude-3-sonnet',
-        messages: [{ role: 'user', content: 'test' }],
-        temperature: 0.5,
-      };
-
-      const result = handlePayload(payload);
-      expect(result).toEqual(payload);
-    });
-  });
-
   describe('Routers Function - Direct Testing', () => {
     it('should generate routers with correct apiTypes', () => {
       const options = { apiKey: 'test', baseURL: 'https://api.newapi.com/v1' };
@@ -823,11 +751,11 @@ describe('NewAPI Runtime - 100% Branch Coverage', () => {
       expect(routers[3].options.baseURL).toBe('https://custom.com/v1');
     });
 
-    it('should configure openai router with handlePayload', () => {
+    it('should configure openai router with useResponseModels', () => {
       const options = { apiKey: 'test', baseURL: 'https://custom.com/v1' };
       const routers = params.routers(options);
 
-      expect((routers[3].options as any).chatCompletion?.handlePayload).toBe(handlePayload);
+      expect((routers[3].options as any).chatCompletion?.useResponseModels).toBeDefined();
     });
 
     it('should filter anthropic models for anthropic router', () => {

packages/model-runtime/src/providers/newapi/index.ts

@@ -4,7 +4,6 @@ import urlJoin from 'url-join';
 import { responsesAPIModels } from '../../const/models';
 import { createRouterRuntime } from '../../core/RouterRuntime';
 import { CreateRouterRuntimeOptions } from '../../core/RouterRuntime/createRuntime';
-import { ChatStreamPayload } from '../../types/chat';
 import { detectModelProvider, processMultiProviderModelList } from '../../utils/modelParse';
 
 export interface NewAPIModelCard {
@@ -26,18 +25,6 @@ export interface NewAPIPricing {
   supported_endpoint_types?: string[];
 }
 
-export const handlePayload = (payload: ChatStreamPayload) => {
-  // Handle OpenAI responses API mode
-  if (
-    responsesAPIModels.has(payload.model) ||
-    payload.model.includes('gpt-') ||
-    /^o\d/.test(payload.model)
-  ) {
-    return { ...payload, apiMode: 'responses' };
-  }
-  return payload;
-};
-
 export const params = {
   debug: {
     chatCompletion: () => process.env.DEBUG_NEWAPI_CHAT_COMPLETION === '1',
@@ -178,7 +165,7 @@ export const params = {
           ...options,
           baseURL: urlJoin(userBaseURL, '/v1'),
           chatCompletion: {
-            handlePayload,
+            useResponseModels: [...Array.from(responsesAPIModels), /gpt-\d(?!\d)/, /^o\d/],
           },
         },
       },

src/app/(backend)/webapi/stt/openai/route.ts

@@ -39,7 +39,10 @@ export const POST = async (req: Request) => {
   // if resOrOpenAI is a Response, it means there is an error,just return it
   if (openaiOrErrResponse instanceof Response) return openaiOrErrResponse;
 
-  const res = await createOpenaiAudioTranscriptions({ openai: openaiOrErrResponse, payload });
+  const res = await createOpenaiAudioTranscriptions({
+    openai: openaiOrErrResponse as any,
+    payload,
+  });
 
   return new Response(JSON.stringify(res), {
     headers: {

src/app/(backend)/webapi/tts/openai/route.ts

@@ -34,5 +34,5 @@ export const POST = async (req: Request) => {
   // if resOrOpenAI is a Response, it means there is an error,just return it
   if (openaiOrErrResponse instanceof Response) return openaiOrErrResponse;
 
-  return await createOpenaiAudioSpeech({ openai: openaiOrErrResponse, payload });
+  return await createOpenaiAudioSpeech({ openai: openaiOrErrResponse as any, payload });
 };

src/app/[variants]/(main)/settings/provider/features/CreateNewProvider/index.tsx

@@ -18,8 +18,8 @@ import { Flexbox } from 'react-layout-kit';
 import { useAiInfraStore } from '@/store/aiInfra/store';
 import { CreateAiProviderParams } from '@/types/aiProvider';
 
-import { CUSTOM_PROVIDER_SDK_OPTIONS } from '../customProviderSdkOptions';
 import { KeyVaultsConfigKey, LLMProviderApiTokenKey, LLMProviderBaseUrlKey } from '../../const';
+import { CUSTOM_PROVIDER_SDK_OPTIONS } from '../customProviderSdkOptions';
 
 interface CreateNewProviderProps {
   onClose?: () => void;
@@ -42,6 +42,15 @@ const CreateNewProvider = memo<CreateNewProviderProps>(({ onClose, open }) => {
         name: values.name || values.id,
       };
 
+      // 只为 openai 和 router (newapi) 类型的自定义 provider 添加 supportResponsesApi: true
+      const sdkType = values.settings?.sdkType;
+      if (sdkType === 'openai' || sdkType === 'router') {
+        finalValues.settings = {
+          ...finalValues.settings,
+          supportResponsesApi: true,
+        };
+      }
+
       await createNewAiProvider(finalValues);
       setLoading(false);
       router.push(`/settings?active=provider&provider=${values.id}`);

src/app/[variants]/(main)/settings/provider/features/customProviderSdkOptions.ts

@@ -9,4 +9,5 @@ export const CUSTOM_PROVIDER_SDK_OPTIONS = [
   { label: 'Qwen', value: 'qwen' },
   { label: 'Volcengine', value: 'volcengine' },
   { label: 'Ollama', value: 'ollama' },
+  { label: 'New API', value: 'router' },
 ] satisfies { label: string; value: AiProviderSDKType }[];

src/config/modelProviders/aihubmix.ts

@@ -11,6 +11,7 @@ const AiHubMix: ModelProviderCard = {
   settings: {
     sdkType: 'router',
     showModelFetcher: true,
+    supportResponsesApi: true,
   },
   url: 'https://aihubmix.com?utm_source=lobehub',
 };

src/config/modelProviders/newapi.ts

@@ -13,6 +13,7 @@ const NewAPI: ModelProviderCard = {
     },
     sdkType: 'router',
     showModelFetcher: true,
+    supportResponsesApi: true,
   },
   url: 'https://github.com/Calcium-Ion/new-api',
 };

src/locales/default/modelProvider.ts

@@ -240,7 +240,7 @@ export default {
       },
       helpDoc: '配置教程',
       responsesApi: {
-        desc: '采用 OpenAI 新一代请求格式规范，解锁思维链等进阶特性',
+        desc: '采用 OpenAI 新一代请求格式规范，解锁思维链等进阶特性 (仅 OpenAI 模型支持)',
         title: '使用 Responses API 规范',
       },
       waitingForMore: '更多模型正在 <1>计划接入</1> 中，敬请期待',