@lobehub/chat 1.49.15 → 1.50.0

package/src/config/aiModels/perplexity.ts

@@ -2,62 +2,55 @@ import { AIChatModelCard } from '@/types/aiModel';
 
 const perplexityChatModels: AIChatModelCard[] = [
   {
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 127_072,
     description:
-      'Llama 3.1 Sonar Small Online 模型，具备8B参数，支持约127,000个标记的上下文长度，专为在线聊天设计，能高效处理各种文本交互。',
-    displayName: 'Llama 3.1 Sonar Small Online',
-    enabled: true,
-    id: 'llama-3.1-sonar-small-128k-online',
-    type: 'chat',
-  },
-  {
-    contextWindowTokens: 128_000,
-    description:
-      'Llama 3.1 Sonar Large Online 模型，具备70B参数，支持约127,000个标记的上下文长度，适用于高容量和多样化聊天任务。',
-    displayName: 'Llama 3.1 Sonar Large Online',
+      '由 DeepSeek 推理模型提供支持的新 API 产品。',
+    displayName: 'Sonar Reasoning',
     enabled: true,
-    id: 'llama-3.1-sonar-large-128k-online',
+    id: 'sonar-reasoning',
     type: 'chat',
   },
   {
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 200_000,
     description:
-      'Llama 3.1 Sonar Huge Online 模型，具备405B参数，支持约127,000个标记的上下文长度，设计用于复杂的在线聊天应用。',
-    displayName: 'Llama 3.1 Sonar Huge Online',
+      '支持搜索上下文的高级搜索产品，支持高级查询和跟进。',
+    displayName: 'Sonar Pro',
     enabled: true,
-    id: 'llama-3.1-sonar-huge-128k-online',
+    id: 'sonar-pro',
     type: 'chat',
   },
   {
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 127_072,
     description:
-      'Llama 3.1 Sonar Small Chat 模型，具备8B参数，专为离线聊天设计，支持约127,000个标记的上下文长度。',
-    displayName: 'Llama 3.1 Sonar Small Chat',
+      '基于搜索上下文的轻量级搜索产品，比 Sonar Pro 更快、更便宜。',
+    displayName: 'Sonar',
     enabled: true,
-    id: 'llama-3.1-sonar-small-128k-chat',
+    id: 'sonar',
     type: 'chat',
   },
+  // The following will be deprecated on 02-22
   {
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 127_072,
     description:
-      'Llama 3.1 Sonar Large Chat 模型，具备70B参数，支持约127,000个标记的上下文长度，适合于复杂的离线聊天任务。',
-    displayName: 'Llama 3.1 Sonar Large Chat',
-    enabled: true,
-    id: 'llama-3.1-sonar-large-128k-chat',
+      'Llama 3.1 Sonar Small Online 模型，具备8B参数，支持约127,000个标记的上下文长度，专为在线聊天设计，能高效处理各种文本交互。',
+    displayName: 'Llama 3.1 Sonar Small Online',
+    id: 'llama-3.1-sonar-small-128k-online',
     type: 'chat',
   },
   {
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 127_072,
     description:
-      'Llama 3.1 8B Instruct 模型，具备8B参数，支持画面指示任务的高效执行，提供优质的文本生成能力。',
-    id: 'llama-3.1-8b-instruct',
+      'Llama 3.1 Sonar Large Online 模型，具备70B参数，支持约127,000个标记的上下文长度，适用于高容量和多样化聊天任务。',
+    displayName: 'Llama 3.1 Sonar Large Online',
+    id: 'llama-3.1-sonar-large-128k-online',
     type: 'chat',
   },
   {
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 127_072,
     description:
-      'Llama 3.1 70B Instruct 模型，具备70B参数，能在大型文本生成和指示任务中提供卓越性能。',
-    id: 'llama-3.1-70b-instruct',
+      'Llama 3.1 Sonar Huge Online 模型，具备405B参数，支持约127,000个标记的上下文长度，设计用于复杂的在线聊天应用。',
+    displayName: 'Llama 3.1 Sonar Huge Online',
+    id: 'llama-3.1-sonar-huge-128k-online',
     type: 'chat',
   },
 ];

package/src/config/modelProviders/perplexity.ts

@@ -4,59 +4,53 @@ import { ModelProviderCard } from '@/types/llm';
 const Perplexity: ModelProviderCard = {
   chatModels: [
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 127_072,
       description:
-        'Llama 3.1 Sonar Small Online 模型，具备8B参数，支持约127,000个标记的上下文长度，专为在线聊天设计，能高效处理各种文本交互。',
-      displayName: 'Llama 3.1 Sonar Small Online',
+        '由 DeepSeek 推理模型提供支持的新 API 产品。',
+      displayName: 'Sonar Reasoning',
       enabled: true,
-      id: 'llama-3.1-sonar-small-128k-online',
+      id: 'sonar-reasoning',
     },
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 200_000,
       description:
-        'Llama 3.1 Sonar Large Online 模型，具备70B参数，支持约127,000个标记的上下文长度，适用于高容量和多样化聊天任务。',
-      displayName: 'Llama 3.1 Sonar Large Online',
+        '支持搜索上下文的高级搜索产品，支持高级查询和跟进。',
+      displayName: 'Sonar Pro',
       enabled: true,
-      id: 'llama-3.1-sonar-large-128k-online',
+      id: 'sonar-pro',
     },
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 127_072,
       description:
-        'Llama 3.1 Sonar Huge Online 模型，具备405B参数，支持约127,000个标记的上下文长度，设计用于复杂的在线聊天应用。',
-      displayName: 'Llama 3.1 Sonar Huge Online',
+        '基于搜索上下文的轻量级搜索产品，比 Sonar Pro 更快、更便宜。',
+      displayName: 'Sonar',
       enabled: true,
-      id: 'llama-3.1-sonar-huge-128k-online',
+      id: 'sonar',
     },
+    // The following will be deprecated on 02-22
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 127_072,
       description:
-        'Llama 3.1 Sonar Small Chat 模型，具备8B参数，专为离线聊天设计，支持约127,000个标记的上下文长度。',
-      displayName: 'Llama 3.1 Sonar Small Chat',
-      enabled: true,
-      id: 'llama-3.1-sonar-small-128k-chat',
-    },
-    {
-      contextWindowTokens: 128_000,
-      description:
-        'Llama 3.1 Sonar Large Chat 模型，具备70B参数，支持约127,000个标记的上下文长度，适合于复杂的离线聊天任务。',
-      displayName: 'Llama 3.1 Sonar Large Chat',
-      enabled: true,
-      id: 'llama-3.1-sonar-large-128k-chat',
+        'Llama 3.1 Sonar Small Online 模型，具备8B参数，支持约127,000个标记的上下文长度，专为在线聊天设计，能高效处理各种文本交互。',
+      displayName: 'Llama 3.1 Sonar Small Online',
+      id: 'llama-3.1-sonar-small-128k-online',
     },
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 127_072,
       description:
-        'Llama 3.1 8B Instruct 模型，具备8B参数，支持画面指示任务的高效执行，提供优质的文本生成能力。',
-      id: 'llama-3.1-8b-instruct',
+        'Llama 3.1 Sonar Large Online 模型，具备70B参数，支持约127,000个标记的上下文长度，适用于高容量和多样化聊天任务。',
+      displayName: 'Llama 3.1 Sonar Large Online',
+      id: 'llama-3.1-sonar-large-128k-online',
     },
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 127_072,
       description:
-        'Llama 3.1 70B Instruct 模型，具备70B参数，能在大型文本生成和指示任务中提供卓越性能。',
-      id: 'llama-3.1-70b-instruct',
+        'Llama 3.1 Sonar Huge Online 模型，具备405B参数，支持约127,000个标记的上下文长度，设计用于复杂的在线聊天应用。',
+      displayName: 'Llama 3.1 Sonar Huge Online',
+      id: 'llama-3.1-sonar-huge-128k-online',
     },
   ],
-  checkModel: 'llama-3.1-8b-instruct',
+  checkModel: 'sonar',
   description:
     'Perplexity 是一家领先的对话生成模型提供商，提供多种先进的Llama 3.1模型，支持在线和离线应用，特别适用于复杂的自然语言处理任务。',
   id: 'perplexity',

package/src/features/AgentSetting/AgentModal/index.tsx

@@ -1,7 +1,7 @@
 'use client';
 
 import { Form, ItemGroup, SliderWithInput } from '@lobehub/ui';
-import { Switch } from 'antd';
+import { Select, Switch } from 'antd';
 import { memo } from 'react';
 import { useTranslation } from 'react-i18next';
 
@@ -17,9 +17,9 @@ const AgentModal = memo(() => {
   const { t } = useTranslation('setting');
   const [form] = Form.useForm();
 
-  const [enableMaxTokens, updateConfig] = useStore((s) => {
+  const [enableMaxTokens, enableReasoningEffort, updateConfig] = useStore((s) => {
     const config = selectors.chatConfig(s);
-    return [config.enableMaxTokens, s.setAgentConfig];
+    return [config.enableMaxTokens, config.enableReasoningEffort, s.setAgentConfig];
   });
 
   const providerName = useProviderName(useStore((s) => s.config.provider) as string);
@@ -79,6 +79,30 @@ const AgentModal = memo(() => {
         name: ['params', 'max_tokens'],
         tag: 'max_tokens',
       },
+      {
+        children: <Switch />,
+        label: t('settingModel.enableReasoningEffort.title'),
+        minWidth: undefined,
+        name: ['chatConfig', 'enableReasoningEffort'],
+        valuePropName: 'checked',
+      },
+      {
+        children: (
+          <Select
+            defaultValue='medium'
+            options={[
+              { label: t('settingModel.reasoningEffort.options.low'), value: 'low' },
+              { label: t('settingModel.reasoningEffort.options.medium'), value: 'medium' },
+              { label: t('settingModel.reasoningEffort.options.high'), value: 'high' },
+            ]}
+          />
+        ),
+        desc: t('settingModel.reasoningEffort.desc'),
+        hidden: !enableReasoningEffort,
+        label: t('settingModel.reasoningEffort.title'),
+        name: ['params', 'reasoning_effort'],
+        tag: 'reasoning_effort',
+      },
     ],
     title: t('settingModel.title'),
   };

package/src/libs/agent-runtime/github/index.ts

@@ -2,7 +2,7 @@ import { LOBE_DEFAULT_MODEL_LIST } from '@/config/modelProviders';
 import type { ChatModelCard } from '@/types/llm';
 
 import { AgentRuntimeErrorType } from '../error';
-import { o1Models, pruneO1Payload } from '../openai';
+import { pruneReasoningPayload, reasoningModels } from '../openai';
 import { ModelProvider } from '../types';
 import {
   CHAT_MODELS_BLOCK_LIST,
@@ -37,8 +37,8 @@ export const LobeGithubAI = LobeOpenAICompatibleFactory({
     handlePayload: (payload) => {
       const { model } = payload;
 
-      if (o1Models.has(model)) {
-        return { ...pruneO1Payload(payload), stream: false } as any;
+      if (reasoningModels.has(model)) {
+        return { ...pruneReasoningPayload(payload), stream: false } as any;
       }
 
       return { ...payload, stream: payload.stream ?? true };

package/src/libs/agent-runtime/openai/index.ts

@@ -2,21 +2,23 @@ import { ChatStreamPayload, ModelProvider, OpenAIChatMessage } from '../types';
 import { LobeOpenAICompatibleFactory } from '../utils/openaiCompatibleFactory';
 
 // TODO: 临时写法，后续要重构成 model card 展示配置
-export const o1Models = new Set([
+export const reasoningModels = new Set([
   'o1-preview',
   'o1-preview-2024-09-12',
   'o1-mini',
   'o1-mini-2024-09-12',
   'o1',
   'o1-2024-12-17',
+  'o3-mini',
+  'o3-mini-2025-01-31',
 ]);
 
-export const pruneO1Payload = (payload: ChatStreamPayload) => ({
+export const pruneReasoningPayload = (payload: ChatStreamPayload) => ({
   ...payload,
   frequency_penalty: 0,
   messages: payload.messages.map((message: OpenAIChatMessage) => ({
     ...message,
-    role: message.role === 'system' ? 'user' : message.role,
+    role: message.role === 'system' ? 'developer' : message.role,
   })),
   presence_penalty: 0,
   temperature: 1,
@@ -29,8 +31,8 @@ export const LobeOpenAI = LobeOpenAICompatibleFactory({
     handlePayload: (payload) => {
       const { model } = payload;
 
-      if (o1Models.has(model)) {
-        return pruneO1Payload(payload) as any;
+      if (reasoningModels.has(model)) {
+        return pruneReasoningPayload(payload) as any;
       }
 
       return { ...payload, stream: payload.stream ?? true };

package/src/libs/agent-runtime/openrouter/__snapshots__/index.test.ts.snap

@@ -196,7 +196,7 @@ _These are extended-context endpoints for [Hermes 3 405B Instruct](/models/nousr
     "contextWindowTokens": 127072,
     "description": "Llama 3.1 Sonar is Perplexity's latest model family. It surpasses their earlier Sonar models in cost-efficiency, speed, and performance. The model is built upon the Llama 3.1 405B and has internet access.",
     "displayName": "Perplexity: Llama 3.1 Sonar 405B Online",
-    "enabled": true,
+    "enabled": false,
     "functionCall": false,
     "id": "perplexity/llama-3.1-sonar-huge-128k-online",
     "maxTokens": undefined,
@@ -304,7 +304,7 @@ Note: This model is experimental and not suited for production use-cases. It may
 
 This is the online version of the [offline chat model](/models/perplexity/llama-3.1-sonar-large-128k-chat). It is focused on delivering helpful, up-to-date, and factual responses. #online",
     "displayName": "Perplexity: Llama 3.1 Sonar 70B Online",
-    "enabled": true,
+    "enabled": false,
     "functionCall": false,
     "id": "perplexity/llama-3.1-sonar-large-128k-online",
     "maxTokens": undefined,
@@ -316,7 +316,7 @@ This is the online version of the [offline chat model](/models/perplexity/llama-
 
 This is a normal offline LLM, but the [online version](/models/perplexity/llama-3.1-sonar-large-128k-online) of this model has Internet access.",
     "displayName": "Perplexity: Llama 3.1 Sonar 70B",
-    "enabled": true,
+    "enabled": false,
     "functionCall": false,
     "id": "perplexity/llama-3.1-sonar-large-128k-chat",
     "maxTokens": undefined,
@@ -328,7 +328,7 @@ This is a normal offline LLM, but the [online version](/models/perplexity/llama-
 
 This is the online version of the [offline chat model](/models/perplexity/llama-3.1-sonar-small-128k-chat). It is focused on delivering helpful, up-to-date, and factual responses. #online",
     "displayName": "Perplexity: Llama 3.1 Sonar 8B Online",
-    "enabled": true,
+    "enabled": false,
     "functionCall": false,
     "id": "perplexity/llama-3.1-sonar-small-128k-online",
     "maxTokens": undefined,
@@ -340,7 +340,7 @@ This is the online version of the [offline chat model](/models/perplexity/llama-
 
 This is a normal offline LLM, but the [online version](/models/perplexity/llama-3.1-sonar-small-128k-online) of this model has Internet access.",
     "displayName": "Perplexity: Llama 3.1 Sonar 8B",
-    "enabled": true,
+    "enabled": false,
     "functionCall": false,
     "id": "perplexity/llama-3.1-sonar-small-128k-chat",
     "maxTokens": undefined,
@@ -354,7 +354,7 @@ It has demonstrated strong performance compared to leading closed-source models
 
 To read more about the model release, [click here](https://ai.meta.com/blog/meta-llama-3/). Usage of this model is subject to [Meta's Acceptable Use Policy](https://llama.meta.com/llama3/use-policy/).",
     "displayName": "Meta: Llama 3.1 70B Instruct",
-    "enabled": false,
+    "enabled": true,
     "functionCall": false,
     "id": "meta-llama/llama-3.1-70b-instruct",
     "maxTokens": undefined,
@@ -384,7 +384,7 @@ It has demonstrated strong performance compared to leading closed-source models
 
 To read more about the model release, [click here](https://ai.meta.com/blog/meta-llama-3/). Usage of this model is subject to [Meta's Acceptable Use Policy](https://llama.meta.com/llama3/use-policy/).",
     "displayName": "Meta: Llama 3.1 8B Instruct",
-    "enabled": false,
+    "enabled": true,
     "functionCall": false,
     "id": "meta-llama/llama-3.1-8b-instruct",
     "maxTokens": undefined,

package/src/libs/agent-runtime/utils/streams/openai.test.ts

@@ -754,6 +754,7 @@ describe('OpenAIStream', () => {
         ].map((i) => `${i}\n`),
       );
     });
+
     it('should handle reasoning in litellm', async () => {
       const data = [
         {
@@ -954,5 +955,206 @@ describe('OpenAIStream', () => {
         ].map((i) => `${i}\n`),
       );
     });
+
+    it('should handle reasoning in siliconflow', async () => {
+      const data = [
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { role: 'assistant', reasoning_content: '', content: '' },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { reasoning_content: '您好', content: '' },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { reasoning_content: '！', content: '' },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '你好', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '很高兴', reasoning_cont: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '为您', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '提供', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '帮助。', reasoning_content: null },
+              logprobs: null,
+              finish_reason: null,
+            },
+          ],
+        },
+        {
+          id: '1',
+          object: 'chat.completion.chunk',
+          created: 1737563070,
+          model: 'deepseek-reasoner',
+          system_fingerprint: 'fp_1c5d8833bc',
+          choices: [
+            {
+              index: 0,
+              delta: { content: '', reasoning_content: null },
+              logprobs: null,
+              finish_reason: 'stop',
+            },
+          ],
+          usage: {
+            prompt_tokens: 6,
+            completion_tokens: 104,
+            total_tokens: 110,
+            prompt_tokens_details: { cached_tokens: 0 },
+            completion_tokens_details: { reasoning_tokens: 70 },
+            prompt_cache_hit_tokens: 0,
+            prompt_cache_miss_tokens: 6,
+          },
+        },
+      ];
+
+      const mockOpenAIStream = new ReadableStream({
+        start(controller) {
+          data.forEach((chunk) => {
+            controller.enqueue(chunk);
+          });
+
+          controller.close();
+        },
+      });
+
+      const protocolStream = OpenAIStream(mockOpenAIStream);
+
+      const decoder = new TextDecoder();
+      const chunks = [];
+
+      // @ts-ignore
+      for await (const chunk of protocolStream) {
+        chunks.push(decoder.decode(chunk, { stream: true }));
+      }
+
+      expect(chunks).toEqual(
+        [
+          'id: 1',
+          'event: reasoning',
+          `data: ""\n`,
+          'id: 1',
+          'event: reasoning',
+          `data: "您好"\n`,
+          'id: 1',
+          'event: reasoning',
+          `data: "！"\n`,
+          'id: 1',
+          'event: text',
+          `data: "你好"\n`,
+          'id: 1',
+          'event: text',
+          `data: "很高兴"\n`,
+          'id: 1',
+          'event: text',
+          `data: "为您"\n`,
+          'id: 1',
+          'event: text',
+          `data: "提供"\n`,
+          'id: 1',
+          'event: text',
+          `data: "帮助。"\n`,
+          'id: 1',
+          'event: stop',
+          `data: "stop"\n`,
+        ].map((i) => `${i}\n`),
+      );
+    });
   });
 });

package/src/libs/agent-runtime/utils/streams/openai.ts

@@ -37,9 +37,8 @@ export const transformOpenAIStream = (
     return { data: errorData, id: 'first_chunk_error', type: 'error' };
   }
 
-  // maybe need another structure to add support for multiple choices
-
   try {
+    // maybe need another structure to add support for multiple choices
     const item = chunk.choices[0];
     if (!item) {
       return { data: chunk, id: chunk.id, type: 'data' };
@@ -88,12 +87,10 @@ export const transformOpenAIStream = (
       return { data: item.finish_reason, id: chunk.id, type: 'stop' };
     }
 
-    if (typeof item.delta?.content === 'string') {
-      return { data: item.delta.content, id: chunk.id, type: 'text' };
-    }
-
-    // DeepSeek reasoner 会将 thinking 放在 reasoning_content 字段中
-    // litellm 处理 reasoning content 时 不会设定 content = null
+    // DeepSeek reasoner will put thinking in the reasoning_content field
+    // litellm will not set content = null when processing reasoning content
+    // en: siliconflow has encountered a situation where both content and reasoning_content are present, so the parsing order go ahead
+    // refs: https://github.com/lobehub/lobe-chat/issues/5681
     if (
       item.delta &&
       'reasoning_content' in item.delta &&
@@ -102,6 +99,10 @@ export const transformOpenAIStream = (
       return { data: item.delta.reasoning_content, id: chunk.id, type: 'reasoning' };
     }
 
+    if (typeof item.delta?.content === 'string') {
+      return { data: item.delta.content, id: chunk.id, type: 'text' };
+    }
+
     // 无内容情况
     if (item.delta && item.delta.content === null) {
       return { data: item.delta, id: chunk.id, type: 'data' };

package/src/locales/default/discover.ts

@@ -127,6 +127,10 @@ export default {
         title: '话题新鲜度',
       },
       range: '范围',
+      reasoning_effort: {
+        desc: '此设置用于控制模型在生成回答前的推理强度。低强度优先响应速度并节省 Token，高强度提供更完整的推理，但会消耗更多 Token 并降低响应速度。默认值为中，平衡推理准确性与响应速度。',
+        title: '推理强度',
+      },
       temperature: {
         desc: '此设置影响模型回应的多样性。较低的值会导致更可预测和典型的回应，而较高的值则鼓励更多样化和不常见的回应。当值设为0时，模型对于给定的输入总是给出相同的回应。',
         title: '随机性',

package/src/locales/default/setting.ts

@@ -202,6 +202,9 @@ export default {
     enableMaxTokens: {
       title: '开启单次回复限制',
     },
+    enableReasoningEffort: {
+      title: '开启推理强度调整',
+    },
     frequencyPenalty: {
       desc: '值越大，越有可能降低重复字词',
       title: '频率惩罚度',
@@ -218,6 +221,15 @@ export default {
       desc: '值越大，越有可能扩展到新话题',
       title: '话题新鲜度',
     },
+    reasoningEffort: {
+      desc: '值越大，推理能力越强，但可能会增加响应时间和 Token 消耗',
+      options: {
+        high: '高',
+        low: '低',
+        medium: '中',
+      },
+      title: '推理强度',
+    },
     temperature: {
       desc: '值越大，回复越随机',
       title: '随机性',

package/src/store/chat/slices/aiChat/actions/generateAIChat.ts

@@ -421,6 +421,11 @@ export const generateAIChat: StateCreator<
       ? agentConfig.params.max_tokens
       : undefined;
 
+    // 5. handle reasoning_effort
+    agentConfig.params.reasoning_effort = chatConfig.enableReasoningEffort
+      ? agentConfig.params.reasoning_effort
+      : undefined;
+
     let isFunctionCall = false;
     let msgTraceId: string | undefined;
     let output = '';