@lobehub/chat 1.113.1 → 1.113.3

package/CHANGELOG.md

@@ -2,6 +2,57 @@
 
 # Changelog
 
+### [Version 1.113.3](https://github.com/lobehub/lobe-chat/compare/v1.113.2...v1.113.3)
+
+<sup>Released on **2025-08-19**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Support Grok thinking models in AiHubMix, The 'stream_options' parameter is only allowed when 'stream' is enabled.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Support Grok thinking models in AiHubMix, closes [#8713](https://github.com/lobehub/lobe-chat/issues/8713) ([ffa9b1b](https://github.com/lobehub/lobe-chat/commit/ffa9b1b))
+- **misc**: The 'stream_options' parameter is only allowed when 'stream' is enabled, closes [#8778](https://github.com/lobehub/lobe-chat/issues/8778) ([fcc32d5](https://github.com/lobehub/lobe-chat/commit/fcc32d5))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
+### [Version 1.113.2](https://github.com/lobehub/lobe-chat/compare/v1.113.1...v1.113.2)
+
+<sup>Released on **2025-08-18**</sup>
+
+#### 🐛 Bug Fixes
+
+- **mcp**: Use customParams for environment settings fallback.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **mcp**: Use customParams for environment settings fallback, closes [#8814](https://github.com/lobehub/lobe-chat/issues/8814) ([ab043d4](https://github.com/lobehub/lobe-chat/commit/ab043d4))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.113.1](https://github.com/lobehub/lobe-chat/compare/v1.113.0...v1.113.1)
 
 <sup>Released on **2025-08-17**</sup>

package/README.md

@@ -255,6 +255,7 @@ We have implemented support for the following model service providers:
 - **[GitHub](https://lobechat.com/discover/provider/github)**: With GitHub Models, developers can become AI engineers and leverage the industry's leading AI models.
 - **[Novita](https://lobechat.com/discover/provider/novita)**: Novita AI is a platform providing a variety of large language models and AI image generation API services, flexible, reliable, and cost-effective. It supports the latest open-source models like Llama3 and Mistral, offering a comprehensive, user-friendly, and auto-scaling API solution for generative AI application development, suitable for the rapid growth of AI startups.
 - **[PPIO](https://lobechat.com/discover/provider/ppio)**: PPIO supports stable and cost-efficient open-source LLM APIs, such as DeepSeek, Llama, Qwen etc.
+- **[302.AI](https://lobechat.com/discover/provider/ai302)**: 302.AI is an on-demand AI application platform offering the most comprehensive AI APIs and online AI applications available on the market.
 - **[Together AI](https://lobechat.com/discover/provider/togetherai)**: Together AI is dedicated to achieving leading performance through innovative AI models, offering extensive customization capabilities, including rapid scaling support and intuitive deployment processes to meet various enterprise needs.
 - **[Fireworks AI](https://lobechat.com/discover/provider/fireworksai)**: Fireworks AI is a leading provider of advanced language model services, focusing on functional calling and multimodal processing. Its latest model, Firefunction V2, is based on Llama-3, optimized for function calling, conversation, and instruction following. The visual language model FireLLaVA-13B supports mixed input of images and text. Other notable models include the Llama series and Mixtral series, providing efficient multilingual instruction following and generation support.
 - **[Groq](https://lobechat.com/discover/provider/groq)**: Groq's LPU inference engine has excelled in the latest independent large language model (LLM) benchmarks, redefining the standards for AI solutions with its remarkable speed and efficiency. Groq represents instant inference speed, demonstrating strong performance in cloud-based deployments.
@@ -283,7 +284,6 @@ We have implemented support for the following model service providers:
 - **[Search1API](https://lobechat.com/discover/provider/search1api)**: Search1API provides access to the DeepSeek series of models that can connect to the internet as needed, including standard and fast versions, supporting a variety of model sizes.
 - **[InfiniAI](https://lobechat.com/discover/provider/infiniai)**: Provides high-performance, easy-to-use, and secure large model services for application developers, covering the entire process from large model development to service deployment.
 - **[Qiniu](https://lobechat.com/discover/provider/qiniu)**: Qiniu, as a long-established cloud service provider, delivers cost-effective and reliable AI inference services for both real-time and batch processing, with a simple and user-friendly experience.
-- **[302.AI](https://lobechat.com/discover/provider/ai302)**: 302.AI is an on-demand AI application platform offering the most comprehensive AI APIs and online AI applications available on the market.
 
 </details>
 

package/README.zh-CN.md

@@ -255,6 +255,7 @@ LobeChat 支持文件上传与知识库功能，你可以上传文件、图片
 - **[GitHub](https://lobechat.com/discover/provider/github)**: 通过 GitHub 模型，开发人员可以成为 AI 工程师，并使用行业领先的 AI 模型进行构建。
 - **[Novita](https://lobechat.com/discover/provider/novita)**: Novita AI 是一个提供多种大语言模型与 AI 图像生成的 API 服务的平台，灵活、可靠且具有成本效益。它支持 Llama3、Mistral 等最新的开源模型，并为生成式 AI 应用开发提供了全面、用户友好且自动扩展的 API 解决方案，适合 AI 初创公司的快速发展。
 - **[PPIO](https://lobechat.com/discover/provider/ppio)**: PPIO 派欧云提供稳定、高性价比的开源模型 API 服务，支持 DeepSeek 全系列、Llama、Qwen 等行业领先大模型。
+- **[302.AI](https://lobechat.com/discover/provider/ai302)**: 302.AI 是一个按需付费的 AI 应用平台，提供市面上最全的 AI API 和 AI 在线应用
 - **[Together AI](https://lobechat.com/discover/provider/togetherai)**: Together AI 致力于通过创新的 AI 模型实现领先的性能，提供广泛的自定义能力，包括快速扩展支持和直观的部署流程，满足企业的各种需求。
 - **[Fireworks AI](https://lobechat.com/discover/provider/fireworksai)**: Fireworks AI 是一家领先的高级语言模型服务商，专注于功能调用和多模态处理。其最新模型 Firefunction V2 基于 Llama-3，优化用于函数调用、对话及指令跟随。视觉语言模型 FireLLaVA-13B 支持图像和文本混合输入。其他 notable 模型包括 Llama 系列和 Mixtral 系列，提供高效的多语言指令跟随与生成支持。
 - **[Groq](https://lobechat.com/discover/provider/groq)**: Groq 的 LPU 推理引擎在最新的独立大语言模型（LLM）基准测试中表现卓越，以其惊人的速度和效率重新定义了 AI 解决方案的标准。Groq 是一种即时推理速度的代表，在基于云的部署中展现了良好的性能。
@@ -283,7 +284,6 @@ LobeChat 支持文件上传与知识库功能，你可以上传文件、图片
 - **[Search1API](https://lobechat.com/discover/provider/search1api)**: Search1API 提供可根据需要自行联网的 DeepSeek 系列模型的访问，包括标准版和快速版本，支持多种参数规模的模型选择。
 - **[InfiniAI](https://lobechat.com/discover/provider/infiniai)**: 为应用开发者提供高性能、易上手、安全可靠的大模型服务，覆盖从大模型开发到大模型服务化部署的全流程。
 - **[Qiniu](https://lobechat.com/discover/provider/qiniu)**: 七牛作为老牌云服务厂商，提供高性价比稳定的实时、批量 AI 推理服务，简单易用。
-- **[302.AI](https://lobechat.com/discover/provider/ai302)**: 302.AI 是一个按需付费的 AI 应用平台，提供市面上最全的 AI API 和 AI 在线应用
 
 </details>
 

package/changelog/v1.json

@@ -1,4 +1,18 @@
 [
+  {
+    "children": {
+      "fixes": [
+        "Support Grok thinking models in AiHubMix, The 'stream_options' parameter is only allowed when 'stream' is enabled."
+      ]
+    },
+    "date": "2025-08-19",
+    "version": "1.113.3"
+  },
+  {
+    "children": {},
+    "date": "2025-08-18",
+    "version": "1.113.2"
+  },
   {
     "children": {},
     "date": "2025-08-17",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.113.1",
+  "version": "1.113.3",
   "description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
   "keywords": [
     "framework",

package/packages/const/src/models.ts

@@ -35,8 +35,6 @@ export const responsesAPIModels = new Set([
   'codex-mini-latest',
   'computer-use-preview',
   'computer-use-preview-2025-03-11',
-  'gpt-5',
-  'gpt-5-mini',
 ]);
 
 /**

package/packages/model-runtime/src/RouterRuntime/baseRuntimeMap.ts

@@ -4,6 +4,7 @@ import { LobeCloudflareAI } from '../cloudflare';
 import { LobeFalAI } from '../fal';
 import { LobeGoogleAI } from '../google';
 import { LobeOpenAI } from '../openai';
+import { LobeXAI } from '../xai';
 
 export const baseRuntimeMap = {
   anthropic: LobeAnthropicAI,
@@ -12,4 +13,5 @@ export const baseRuntimeMap = {
   fal: LobeFalAI,
   google: LobeGoogleAI,
   openai: LobeOpenAI,
+  xai: LobeXAI,
 };

package/packages/model-runtime/src/aihubmix/index.ts

@@ -65,6 +65,13 @@ export const LobeAiHubMixAI = createRouterRuntime({
       ),
       options: { baseURL: urlJoin(baseURL, '/gemini') },
     },
+    {
+      apiType: 'xai',
+      models: LOBE_DEFAULT_MODEL_LIST.map((m) => m.id).filter(
+        (id) => detectModelProvider(id) === 'xai',
+      ),
+      options: { baseURL: urlJoin(baseURL, '/v1') },
+    },
     {
       apiType: 'openai',
       options: {

package/packages/model-runtime/src/openai/index.ts

@@ -11,6 +11,16 @@ export interface OpenAIModelCard {
 
 const prunePrefixes = ['o1', 'o3', 'o4', 'codex', 'computer-use', 'gpt-5'];
 const oaiSearchContextSize = process.env.OPENAI_SEARCH_CONTEXT_SIZE; // low, medium, high
+const enableServiceTierFlex = process.env.OPENAI_SERVICE_TIER_FLEX === '1';
+const flexSupportedModels = ['gpt-5', 'o3', 'o4-mini']; // Flex 处理仅适用于这些模型
+
+const supportsFlexTier = (model: string) => {
+  // 排除 o3-mini，其不支持 Flex 处理
+  if (model.startsWith('o3-mini')) {
+    return false;
+  }
+  return flexSupportedModels.some(supportedModel => model.startsWith(supportedModel));
+};
 
 export const LobeOpenAI = createOpenAICompatibleRuntime({
   baseURL: 'https://api.openai.com/v1',
@@ -32,6 +42,7 @@ export const LobeOpenAI = createOpenAICompatibleRuntime({
           frequency_penalty: undefined,
           model,
           presence_penalty: undefined,
+          ...(enableServiceTierFlex && supportsFlexTier(model) && { service_tier: 'flex' }),
           stream: payload.stream ?? true,
           temperature: undefined,
           top_p: undefined,
@@ -43,7 +54,7 @@ export const LobeOpenAI = createOpenAICompatibleRuntime({
         } as any;
       }
 
-      return { ...rest, model, stream: payload.stream ?? true };
+      return { ...rest, model, ...(enableServiceTierFlex && supportsFlexTier(model) && { service_tier: 'flex' }), stream: payload.stream ?? true };
     },
   },
   debug: {
@@ -81,6 +92,7 @@ export const LobeOpenAI = createOpenAICompatibleRuntime({
           reasoning: payload.reasoning
             ? { ...payload.reasoning, summary: 'auto' }
             : { summary: 'auto' },
+          ...(enableServiceTierFlex && supportsFlexTier(model) && { service_tier: 'flex' }),
           stream: payload.stream ?? true,
           tools: openaiTools as any,
           // computer-use series must set truncation as auto
@@ -91,7 +103,7 @@ export const LobeOpenAI = createOpenAICompatibleRuntime({
         }) as any;
       }
 
-      return { ...rest, model, stream: payload.stream ?? true, tools: openaiTools } as any;
+      return { ...rest, model, ...(enableServiceTierFlex && supportsFlexTier(model) && { service_tier: 'flex' }), stream: payload.stream ?? true, tools: openaiTools } as any;
     },
   },
 });

package/packages/model-runtime/src/utils/openaiHelpers.ts

@@ -101,8 +101,11 @@ export const convertOpenAIResponseInputs = async (
 };
 
 export const pruneReasoningPayload = (payload: ChatStreamPayload) => {
+  const shouldStream = !disableStreamModels.has(payload.model);
+  const { stream_options, ...cleanedPayload } = payload as any;
+  
   return {
-    ...payload,
+    ...cleanedPayload,
     frequency_penalty: 0,
     messages: payload.messages.map((message: OpenAIChatMessage) => ({
       ...message,
@@ -114,7 +117,9 @@ export const pruneReasoningPayload = (payload: ChatStreamPayload) => {
           : message.role,
     })),
     presence_penalty: 0,
-    stream: !disableStreamModels.has(payload.model),
+    stream: shouldStream,
+    // Only include stream_options when stream is enabled
+    ...(shouldStream && stream_options && { stream_options }),
     temperature: 1,
     top_p: 1,
   };

package/src/config/aiModels/groq.ts

@@ -33,7 +33,7 @@ const groqChatModels: AIChatModelCard[] = [
       'OpenAI GPT-OSS 120B 是一款拥有 1200 亿参数的顶尖语言模型，内置浏览器搜索和代码执行功能，并具备推理能力。',
     displayName: 'GPT OSS 120B',
     id: 'openai/gpt-oss-120b',
-    maxOutput: 32_768,
+    maxOutput: 65_536,
     pricing: {
       units: [
         { name: 'textInput', rate: 0.15, strategy: 'fixed', unit: 'millionTokens' },
@@ -53,7 +53,7 @@ const groqChatModels: AIChatModelCard[] = [
       'OpenAI GPT-OSS 20B 是一款拥有 200 亿参数的顶尖语言模型，内置浏览器搜索和代码执行功能，并具备推理能力。',
     displayName: 'GPT OSS 20B',
     id: 'openai/gpt-oss-20b',
-    maxOutput: 32_768,
+    maxOutput: 65_536,
     pricing: {
       units: [
         { name: 'textInput', rate: 0.1, strategy: 'fixed', unit: 'millionTokens' },

package/src/config/aiModels/qwen.ts

@@ -519,21 +519,61 @@ const qwenChatModels: AIChatModelCard[] = [
       search: true,
     },
     config: {
-      deploymentName: 'qwen-plus-2025-07-14',
+      deploymentName: 'qwen-plus-2025-07-28',
     },
-    contextWindowTokens: 131_072,
+    contextWindowTokens: 1_000_000,
     description: '通义千问超大规模语言模型增强版，支持中文、英文等不同语言输入。',
     displayName: 'Qwen Plus',
     enabled: true,
     id: 'qwen-plus',
-    maxOutput: 16_384,
+    maxOutput: 32_768,
     organization: 'Qwen',
     pricing: {
       currency: 'CNY',
       units: [
-        { name: 'textInput_cacheRead', rate: 0.32, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textInput', rate: 0.8, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 8, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          lookup: {
+            prices: {
+              '[0, 128_000]': 0.8 * 0.4,
+              '[128_000, 256_000]': 2.4 * 0.4,
+              '[256_000, infinity]': 4.8 * 0.4,
+            },
+            pricingParams: ['textInputRange'],
+          },
+          name: 'textInput_cacheRead',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
+        {
+          lookup: {
+            prices: {
+              '[0, 128_000]': 0.8,
+              '[128_000, 256_000]': 2.4,
+              '[256_000, infinity]': 4.8,
+            },
+            pricingParams: ['textInputRange'],
+          },
+          name: 'textInput',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
+        {
+          lookup: {
+            prices: {
+              '[0, 128_000]_[false]': 2,
+              '[0, 128_000]_[true]': 8,
+              '[128_000, 256_000]_[false]': 20,
+
+              '[128_000, 256_000]_[true]': 24,
+              '[256_000, infinity]_[false]': 48,
+              '[256_000, infinity]_[true]': 64,
+            },
+            pricingParams: ['textInputRange', 'thinkingMode'],
+          },
+          name: 'textOutput',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
       ],
     },
     releasedAt: '2025-07-14',
@@ -643,7 +683,7 @@ const qwenChatModels: AIChatModelCard[] = [
       vision: true,
     },
     config: {
-      deploymentName: 'qwen-vl-plus-2025-01-25',
+      deploymentName: 'qwen-vl-plus-2025-08-15',
     },
     contextWindowTokens: 131_072,
     description:
@@ -655,9 +695,9 @@ const qwenChatModels: AIChatModelCard[] = [
     pricing: {
       currency: 'CNY',
       units: [
-        { name: 'textInput_cacheRead', rate: 0.6, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textInput', rate: 1.5, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 4.5, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput_cacheRead', rate: 0.8 * 0.4, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput', rate: 0.8, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 2, strategy: 'fixed', unit: 'millionTokens' },
       ],
     },
     type: 'chat',
@@ -667,7 +707,7 @@ const qwenChatModels: AIChatModelCard[] = [
       vision: true,
     },
     config: {
-      deploymentName: 'qwen-vl-max-2025-04-08',
+      deploymentName: 'qwen-vl-max-2025-08-13',
     },
     contextWindowTokens: 131_072,
     description:
@@ -679,9 +719,9 @@ const qwenChatModels: AIChatModelCard[] = [
     pricing: {
       currency: 'CNY',
       units: [
-        { name: 'textInput_cacheRead', rate: 1.2, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textInput', rate: 3, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 9, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput_cacheRead', rate: 1.6 * 0.4, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput', rate: 1.6, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
       ],
     },
     type: 'chat',
@@ -1304,6 +1344,30 @@ const qwenChatModels: AIChatModelCard[] = [
 ];
 
 const qwenImageModels: AIImageModelCard[] = [
+  {
+    description:
+      'Qwen-Image 是一款通用图像生成模型，支持多种艺术风格，尤其擅长复杂文本渲染，特别是中英文文本渲染。模型支持多行布局、段落级文本生成以及细粒度细节刻画，可实现复杂的图文混合布局设计。',
+    displayName: 'Qwen Image',
+    enabled: true,
+    id: 'qwen-image',
+    organization: 'Qwen',
+    parameters: {
+      prompt: {
+        default: '',
+      },
+      seed: { default: null },
+      size: {
+        default: '1328*1328',
+        enum: ['1664*928', '1472*1140', '1328*1328', '1140*1472', '928*1664'],
+      },
+    },
+    pricing: {
+      currency: 'CNY',
+      units: [{ name: 'imageGeneration', rate: 0.25, strategy: 'fixed', unit: 'image' }],
+    },
+    releasedAt: '2025-08-13',
+    type: 'image',
+  },
   {
     description:
       '万相2.2极速版，当前最新模型。在创意性、稳定性、写实质感上全面升级，生成速度快，性价比高。',

package/src/config/aiModels/siliconcloud.ts

@@ -61,6 +61,27 @@ const siliconcloudChatModels: AIChatModelCard[] = [
     releasedAt: '2025-07-31',
     type: 'chat',
   },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      vision: true,
+    },
+    contextWindowTokens: 65_536,
+    description:
+      'GLM-4.5V 是由智谱 AI（Zhipu AI）发布的最新一代视觉语言模型（VLM）该模型基于拥有 106B 总参数和 12B 激活参数的旗舰文本模型 GLM-4.5-Air 构建，采用了混合专家（MoE）架构，旨在以更低的推理成本实现卓越性能 GLM-4.5V 在技术上延续了 GLM-4.1V-Thinking 的路线，并引入了三维旋转位置编码（3D-RoPE）等创新，显著增强了对三维空间关系的感知与推理能力。通过在预训练、监督微调和强化学习阶段的优化，该模型具备了处理图像、视频、长文档等多种视觉内容的能力，在 41 个公开的多模态基准测试中达到了同级别开源模型的顶尖水平此外，模型还新增了“思考模式”开关，允许用户在快速响应和深度推理之间灵活选择，以平衡效率与效果。',
+    displayName: 'GLM-4.5V',
+    id: 'zai-org/GLM-4.5V',
+    pricing: {
+      currency: 'CNY',
+      units: [
+        { name: 'textInput', rate: 1, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 6, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    releasedAt: '2025-08-11',
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,

package/src/config/modelProviders/index.ts

@@ -136,6 +136,7 @@ export const DEFAULT_MODEL_PROVIDER_LIST = [
   BflProvider,
   NovitaProvider,
   PPIOProvider,
+  Ai302Provider,
   NvidiaProvider,
   TogetherAIProvider,
   FireworksAIProvider,
@@ -172,7 +173,6 @@ export const DEFAULT_MODEL_PROVIDER_LIST = [
   Search1APIProvider,
   InfiniAIProvider,
   QiniuProvider,
-  Ai302Provider,
 ];
 
 export const filterEnabledModels = (provider: ModelProviderCard) => {

package/src/services/discover.ts

@@ -143,6 +143,10 @@ class DiscoverService {
     errorCode,
     ...params
   }: InstallReportRequest) => {
+    // if user don't allow tracing, just not report installation
+    const allow = preferenceSelectors.userAllowTrace(useUserStore.getState());
+
+    if (!allow) return;
     await this.injectMPToken();
 
     const reportData = {
@@ -164,7 +168,8 @@ class DiscoverService {
    * 上报插件调用结果
    */
   reportPluginCall = async (reportData: CallReportRequest) => {
-    const allow = useUserStore(preferenceSelectors.userAllowTrace);
+    // if user don't allow tracing , just not report calling
+    const allow = preferenceSelectors.userAllowTrace(useUserStore.getState());
 
     if (!allow) return;
 

package/src/services/mcp.ts

@@ -45,7 +45,7 @@ class MCPService {
 
     const data = {
       args,
-      env: plugin.settings,
+      env: plugin.settings || plugin.customParams?.mcp?.env,
       params: { ...plugin.customParams?.mcp, name: identifier } as any,
       toolName: apiName,
     };