@lobehub/chat 1.50.3 → 1.50.5

package/.env.example

@@ -106,6 +106,7 @@ OPENAI_API_KEY=sk-xxxxxxxxx
 
 ### DeepSeek AI  ####
 
+# DEEPSEEK_PROXY_URL=https://api.deepseek.com/v1
 # DEEPSEEK_API_KEY=xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
 
 ### Qwen AI  ####

package/CHANGELOG.md

@@ -2,6 +2,57 @@
 
 # Changelog
 
+### [Version 1.50.5](https://github.com/lobehub/lobe-chat/compare/v1.50.4...v1.50.5)
+
+<sup>Released on **2025-02-04**</sup>
+
+#### 💄 Styles
+
+- **misc**: Add/Update Aliyun Cloud Models, update GitHub Models.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### Styles
+
+- **misc**: Add/Update Aliyun Cloud Models, closes [#5613](https://github.com/lobehub/lobe-chat/issues/5613) ([95cd822](https://github.com/lobehub/lobe-chat/commit/95cd822))
+- **misc**: Update GitHub Models, closes [#5683](https://github.com/lobehub/lobe-chat/issues/5683) ([ed4e048](https://github.com/lobehub/lobe-chat/commit/ed4e048))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
+### [Version 1.50.4](https://github.com/lobehub/lobe-chat/compare/v1.50.3...v1.50.4)
+
+<sup>Released on **2025-02-04**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Fix invalid utf8 character.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Fix invalid utf8 character, closes [#5732](https://github.com/lobehub/lobe-chat/issues/5732) ([2905cb5](https://github.com/lobehub/lobe-chat/commit/2905cb5))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.50.3](https://github.com/lobehub/lobe-chat/compare/v1.50.2...v1.50.3)
 
 <sup>Released on **2025-02-04**</sup>

package/README.ja-JP.md

@@ -302,14 +302,14 @@ LobeChat エージェントマーケットプレイスでは、クリエイタ
 
 <!-- AGENT LIST -->
 
-| 最近追加                                                                                                                                                                     | 説明                                                                                                                              |
-| ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------------------- |
-| [SUNO 曲作支援ツール](https://lobechat.com/discover/assistant/suno-lyrics-assistant)<br/><sup>By **[sqkkyzx](https://github.com/sqkkyzx)** on **2025-01-26**</sup>           | ユーザーのニーズに基づいて SUNO の曲作成パラメータを生成します<br/>`歌詞作成` `音楽スタイル` `編曲` `パラメータ設定`              |
-| [偉大なるビッグス・ディッカス](https://lobechat.com/discover/assistant/all-knowing)<br/><sup>By **[CGitwater](https://github.com/CGitwater)** on **2025-01-24**</sup>        | 全能の知識の神<br/>`ビッグス` `ディッカス`                                                                                        |
-| [PPT 制作達人](https://lobechat.com/discover/assistant/ppt-production-expert)<br/><sup>By **[patricleehua](https://github.com/patricleehua)** on **2025-01-24**</sup>        | 高品質な PPT の迅速な制作と最適化に優れています<br/>`ppt制作` `デザイン` `コンサルティング` `コンテンツ最適化` `ユーザーサポート` |
-| [OCR ドキュメント転写アシスタント](https://lobechat.com/discover/assistant/ocr-markdown)<br/><sup>By **[Liangpi000](https://github.com/Liangpi000)** on **2025-01-24**</sup> | 文書内容の転写と markdown フォーマットに優れています<br/>`文書生成` `markdown` `フォーマット` `転写` `タスクガイド`               |
-
-> 📊 Total agents: [<kbd>**471**</kbd> ](https://lobechat.com/discover/assistants)
+| 最近追加                                                                                                                                                     | 説明                                                                                                                                    |
+| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------- |
+| [鋭い評論家](https://lobechat.com/discover/assistant/ruipingshi)<br/><sup>By **[Zippland](https://github.com/Zippland)** on **2025-02-04**</sup>             | 鋭い評論と深い問題分析が得意<br/>`評論` `社会的見解` `鋭い分析`                                                                         |
+| [Python の天才](https://lobechat.com/discover/assistant/python-genius)<br/><sup>By **[novaspivack](https://github.com/novaspivack)** on **2025-02-04**</sup> | 高度な Python コーダー<br/>`コード` `python`                                                                                            |
+| [SAT マスター](https://lobechat.com/discover/assistant/sat-teaching)<br/><sup>By **[iBz-04](https://github.com/iBz-04)** on **2025-02-04**</sup>             | 1300 点以上のスコアを目指すデジタル SAT コーチングの専門家<br/>`sat` `適性試験`                                                         |
+| [宇宙の啓示者](https://lobechat.com/discover/assistant/universal-god)<br/><sup>By **[GowayLee](https://github.com/GowayLee)** on **2025-02-04**</sup>        | 時空を超えた知恵の神託、生命の本質を洞察する<br/>`キャラクターデザイン` `aiキャラクター` `メタバース` `ロールプレイング` `知恵システム` |
+
+> 📊 Total agents: [<kbd>**478**</kbd> ](https://lobechat.com/discover/assistants)
 
  <!-- AGENT LIST -->
 

package/README.md

@@ -319,14 +319,14 @@ Our marketplace is not just a showcase platform but also a collaborative space.
 
 <!-- AGENT LIST -->
 
-| Recent Submits                                                                                                                                                                   | Description                                                                                                                                             |
-| -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------- |
-| [SUNO Song Creation Assistant](https://lobechat.com/discover/assistant/suno-lyrics-assistant)<br/><sup>By **[sqkkyzx](https://github.com/sqkkyzx)** on **2025-01-26**</sup>      | Can generate SUNO song creation parameters based on user needs<br/>`lyric-creation` `music-style` `arrangement` `parameter-settings`                    |
-| [The Great Biggus Dickus](https://lobechat.com/discover/assistant/all-knowing)<br/><sup>By **[CGitwater](https://github.com/CGitwater)** on **2025-01-24**</sup>                 | The almighty powerful god of klnowledge<br/>`biggus` `diccus`                                                                                           |
-| [PPT Production Expert](https://lobechat.com/discover/assistant/ppt-production-expert)<br/><sup>By **[patricleehua](https://github.com/patricleehua)** on **2025-01-24**</sup>   | Skilled in the rapid production and optimization of high-quality PPTs<br/>`ppt-production` `design` `consulting` `content-optimization` `user-support`  |
-| [OCR Document Transcription Assistant](https://lobechat.com/discover/assistant/ocr-markdown)<br/><sup>By **[Liangpi000](https://github.com/Liangpi000)** on **2025-01-24**</sup> | Specializes in document content transcription and markdown formatting<br/>`document-generation` `markdown` `formatting` `transcription` `task-guidance` |
-
-> 📊 Total agents: [<kbd>**471**</kbd> ](https://lobechat.com/discover/assistants)
+| Recent Submits                                                                                                                                               | Description                                                                                                                                           |
+| ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------- |
+| [Sharp Commentator](https://lobechat.com/discover/assistant/ruipingshi)<br/><sup>By **[Zippland](https://github.com/Zippland)** on **2025-02-04**</sup>      | Specializes in sharp commentary and in-depth analysis of issues<br/>`commentary` `social-perspectives` `sharp-analysis`                               |
+| [Python Genius](https://lobechat.com/discover/assistant/python-genius)<br/><sup>By **[novaspivack](https://github.com/novaspivack)** on **2025-02-04**</sup> | An advanced python coder<br/>`code` `python`                                                                                                          |
+| [SAT master](https://lobechat.com/discover/assistant/sat-teaching)<br/><sup>By **[iBz-04](https://github.com/iBz-04)** on **2025-02-04**</sup>               | Expert in Digital SAT coaching for 1300+ scores<br/>`sat` `aptitude-test`                                                                             |
+| [Cosmic Oracle](https://lobechat.com/discover/assistant/universal-god)<br/><sup>By **[GowayLee](https://github.com/GowayLee)** on **2025-02-04**</sup>       | Wisdom from across time and space, insight into the essence of life<br/>`character-design` `ai-characters` `metaverse` `role-playing` `wisdom-system` |
+
+> 📊 Total agents: [<kbd>**478**</kbd> ](https://lobechat.com/discover/assistants)
 
  <!-- AGENT LIST -->
 

package/README.zh-CN.md

@@ -308,14 +308,14 @@ LobeChat 的插件生态系统是其核心功能的重要扩展，它极大地
 
 <!-- AGENT LIST -->
 
-| 最近新增                                                                                                                                                              | 描述                                                                                  |
-| --------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
-| [SUNO 歌曲创作助手](https://lobechat.com/discover/assistant/suno-lyrics-assistant)<br/><sup>By **[sqkkyzx](https://github.com/sqkkyzx)** on **2025-01-26**</sup>      | 能根据用户需求生成 SUNO 歌曲创作参数<br/>`歌词创作` `音乐风格` `编曲` `参数设置`      |
-| [伟大的比古斯・迪克斯](https://lobechat.com/discover/assistant/all-knowing)<br/><sup>By **[CGitwater](https://github.com/CGitwater)** on **2025-01-24**</sup>         | 全能强大的知识之神<br/>`比古斯` `迪克斯`                                              |
-| [PPT 制作达人](https://lobechat.com/discover/assistant/ppt-production-expert)<br/><sup>By **[patricleehua](https://github.com/patricleehua)** on **2025-01-24**</sup> | 擅长高质量 PPT 的快速制作和优化<br/>`ppt制作` `设计` `咨询` `内容优化` `用户支持`     |
-| [OCR 文档转录助手](https://lobechat.com/discover/assistant/ocr-markdown)<br/><sup>By **[Liangpi000](https://github.com/Liangpi000)** on **2025-01-24**</sup>          | 擅长文件内容转录与 markdown 格式<br/>`文档生成` `markdown` `格式化` `转录` `任务指导` |
-
-> 📊 Total agents: [<kbd>**471**</kbd> ](https://lobechat.com/discover/assistants)
+| 最近新增                                                                                                                                                   | 描述                                                                                  |
+| ---------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------- |
+| [锐评师](https://lobechat.com/discover/assistant/ruipingshi)<br/><sup>By **[Zippland](https://github.com/Zippland)** on **2025-02-04**</sup>               | 擅长犀利点评与深度剖析问题<br/>`评论` `社会观点` `尖锐分析`                           |
+| [Python 天才](https://lobechat.com/discover/assistant/python-genius)<br/><sup>By **[novaspivack](https://github.com/novaspivack)** on **2025-02-04**</sup> | 一名高级 Python 编程者<br/>`代码` `python`                                            |
+| [SAT 大师](https://lobechat.com/discover/assistant/sat-teaching)<br/><sup>By **[iBz-04](https://github.com/iBz-04)** on **2025-02-04**</sup>               | 数字 SAT 辅导专家，帮助学生取得 1300 + 分数<br/>`sat` `能力测试`                      |
+| [宇宙启示者](https://lobechat.com/discover/assistant/universal-god)<br/><sup>By **[GowayLee](https://github.com/GowayLee)** on **2025-02-04**</sup>        | 跨时空的智慧神谕，洞悉生命本质<br/>`角色设计` `ai角色` `元宇宙` `角色扮演` `智慧系统` |
+
+> 📊 Total agents: [<kbd>**478**</kbd> ](https://lobechat.com/discover/assistants)
 
  <!-- AGENT LIST -->
 

package/changelog/v1.json

@@ -1,4 +1,22 @@
 [
+  {
+    "children": {
+      "improvements": [
+        "Add/Update Aliyun Cloud Models, update GitHub Models."
+      ]
+    },
+    "date": "2025-02-04",
+    "version": "1.50.5"
+  },
+  {
+    "children": {
+      "fixes": [
+        "Fix invalid utf8 character."
+      ]
+    },
+    "date": "2025-02-04",
+    "version": "1.50.4"
+  },
   {
     "children": {
       "improvements": [

package/docs/changelog/2025-01-22-new-ai-provider.mdx

@@ -1,8 +1,8 @@
 ---
 title: LobeChat Launches New AI Provider Management System
 description: >-
-  LobeChat has revamped its AI Provider Management System, now supporting custom AI providers and models.
-
+  LobeChat has revamped its AI Provider Management System, now supporting custom
+  AI providers and models.
 tags:
   - LobeChat
   - AI Provider

package/docs/changelog/2025-02-02-deepseek-r1.mdx

@@ -0,0 +1,33 @@
+---
+title: >-
+  LobeChat Integrates DeepSeek R1, Bringing a Revolutionary Chain of Thought Experience
+
+description: >-
+  LobeChat v1.49.12 fully supports the DeepSeek R1 model, providing users with an unprecedented interactive experience in the chain of thought.
+
+tags:
+  - LobeChat
+  - DeepSeek
+  - Chain of Thought
+---
+
+# Perfect Integration of DeepSeek R1 and it's Deep Thinking Experience 🎉
+
+After nearly 10 days of meticulous refinement, LobeChat has fully integrated the DeepSeek R1 model in version v1.49.12, offering users a revolutionary interactive experience in the chain of thought!
+
+## 🚀 Major Updates
+
+- 🤯 **Comprehensive Support for DeepSeek R1**: Now fully integrated in both the Community and Cloud versions ([lobechat.com](https://lobechat.com)).
+- 🧠 **Real-Time Chain of Thought Display**: Transparently presents the AI's reasoning process, making the resolution of complex issues clear and visible.
+- ⚡️ **Deep Thinking Experience**: Utilizing Chain of Thought technology, it provides more insightful AI conversations.
+- 💫 **Intuitive Problem Analysis**: Makes the analysis of complex issues clear and easy to understand.
+
+## 🌟 How to Use
+
+1. Upgrade to LobeChat v1.49.12 or visit [lobechat.com](https://lobechat.com).
+2. Select the DeepSeek R1 model in the settings.
+3. Experience a whole new level of intelligent conversation!
+
+## 📢 Feedback and Support
+
+If you encounter any issues while using the application or have suggestions for new features, feel free to engage with us through GitHub Discussions. Let's work together to create a better LobeChat!

package/docs/changelog/2025-02-02-deepseek-r1.zh-CN.mdx

@@ -0,0 +1,29 @@
+---
+title: LobeChat 重磅集成 DeepSeek R1，带来革命性思维链体验
+description: LobeChat v1.49.12 已完整支持 DeepSeek R1 模型，为用户带来前所未有的思维链交互体验
+tags:
+  - DeepSeek R1
+  - CoT
+  - 思维链
+---
+
+# 完美集成 DeepSeek R1 ，开启思维链新体验
+
+经过近 10 天的精心打磨，LobeChat 已在 v1.49.12 版本中完整集成了 DeepSeek R1 模型，为用户带来革命性的思维链交互体验！
+
+## 🚀 重大更新
+
+- 🤯 **DeepSeek R1 全面支持**: 现已在社区版与 Cloud 版（[lobechat.com](https://lobechat.com)）中完整接入
+- 🧠 **实时思维链展示**: 透明呈现 AI 的推理过程，让复杂问题的解决过程清晰可见
+- ⚡️ **深度思考体验**: 通过 Chain of Thought 技术，带来更具洞察力的 AI 对话
+- 💫 **直观的问题解析**: 让复杂问题的分析过程变得清晰易懂
+
+## 🌟 使用方式
+
+1. 升级到 LobeChat v1.49.12 或访问 [lobechat.com](https://lobechat.com)
+2. 在设置中选择 DeepSeek R1 模型
+3. 开启全新的智能对话体验！
+
+## 📢 反馈与支持
+
+如果您在使用过程中遇到任何问题，或对新功能有任何建议，欢迎通过 GitHub Discussions 与我们交流。让我们一起打造更好的 LobeChat！

package/docs/changelog/index.json

@@ -2,6 +2,12 @@
   "$schema": "https://github.com/lobehub/lobe-chat/blob/main/docs/changelog/schema.json",
   "cloud": [],
   "community": [
+    {
+      "image": "https://github.com/user-attachments/assets/5fe4c373-ebd0-42a9-bdca-0ab7e0a2e747",
+      "id": "2025-02-02-deepseek-r1",
+      "date": "2025-02-02",
+      "versionRange": ["1.47.8", "1.49.12"]
+    },
     {
       "image": "https://github.com/user-attachments/assets/7350f211-61ce-488e-b0e2-f0fcac25caeb",
       "id": "2025-01-22-new-ai-provider",

package/docs/self-hosting/environment-variables/model-provider.mdx

@@ -169,6 +169,13 @@ If you need to use Azure OpenAI to provide model services, you can refer to the
 
 ## DeepSeek AI
 
+### `DEEPSEEK_PROXY_URL`
+
+- Type: Optional
+- Description: If you manually configure the DeepSeek API proxy, you can use this configuration item to override the default DeepSeek API request base URL
+- Default: -
+- Example: `https://xxxx.models.ai.azure.com/v1`
+
 ### `DEEPSEEK_API_KEY`
 
 - Type: Required

package/docs/self-hosting/environment-variables/model-provider.zh-CN.mdx

@@ -167,6 +167,13 @@ LobeChat 在部署时提供了丰富的模型服务商相关的环境变量，
 
 ## DeepSeek AI
 
+### `DEEPSEEK_PROXY_URL`
+
+- 类型：可选
+- 描述：如果您手动配置了 DeepSeek API 代理，可以使用此配置项覆盖默认的 DeepSeek API 请求基础 URL
+- 默认值: -
+- 示例: `https://xxxx.models.ai.azure.com/v1`
+
 ### `DEEPSEEK_API_KEY`
 
 - 类型：必选

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.50.3",
+  "version": "1.50.5",
   "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/src/config/aiModels/github.ts

@@ -58,7 +58,7 @@ const githubChatModels: AIChatModelCard[] = [
       functionCall: true,
       vision: true,
     },
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 134_144,
     description: '一种经济高效的AI解决方案，适用于多种文本和图像任务。',
     displayName: 'OpenAI GPT-4o mini',
     enabled: true,
@@ -71,11 +71,18 @@ const githubChatModels: AIChatModelCard[] = [
       functionCall: true,
       vision: true,
     },
-    contextWindowTokens: 128_000,
+    contextWindowTokens: 134_144,
     description: 'OpenAI GPT-4系列中最先进的多模态模型，可以处理文本和图像输入。',
     displayName: 'OpenAI GPT-4o',
     enabled: true,
     id: 'gpt-4o',
+    maxOutput: 16_384,
+    type: 'chat',
+  },
+  {
+    contextWindowTokens: 128_000,
+    displayName: 'DeepSeek R1',
+    id: 'DeepSeek-R1',
     maxOutput: 4096,
     type: 'chat',
   },
@@ -145,6 +152,13 @@ const githubChatModels: AIChatModelCard[] = [
     maxOutput: 4096,
     type: 'chat',
   },
+  {
+    contextWindowTokens: 262_144,
+    displayName: 'Codestral',
+    id: 'Codestral-2501',
+    maxOutput: 4096,
+    type: 'chat',
+  },
   {
     abilities: {
       vision: true,
@@ -210,6 +224,20 @@ const githubChatModels: AIChatModelCard[] = [
     maxOutput: 4096,
     type: 'chat',
   },
+  {
+    contextWindowTokens: 16_384,
+    displayName: 'Phi 4',
+    id: 'Phi-4',
+    maxOutput: 16_384,
+    type: 'chat',
+  },
+  {
+    contextWindowTokens: 131_072,
+    displayName: 'Phi 3.5 MoE',
+    id: 'Phi-3.5-MoE-instruct',
+    maxOutput: 4096,
+    type: 'chat',
+  },
   {
     contextWindowTokens: 131_072,
     description: 'Phi-3-mini模型的更新版。',

package/src/config/aiModels/qwen.ts

@@ -5,11 +5,12 @@ const qwenChatModels: AIChatModelCard[] = [
     abilities: {
       functionCall: true,
     },
-    contextWindowTokens: 131_072,
+    contextWindowTokens: 1_000_000,
     description: '通义千问超大规模语言模型，支持中文、英文等不同语言输入。',
     displayName: 'Qwen Turbo',
     enabled: true,
     id: 'qwen-turbo-latest',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 0.3,
@@ -26,6 +27,7 @@ const qwenChatModels: AIChatModelCard[] = [
     displayName: 'Qwen Plus',
     enabled: true,
     id: 'qwen-plus-latest',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 0.8,
@@ -43,6 +45,7 @@ const qwenChatModels: AIChatModelCard[] = [
     displayName: 'Qwen Max',
     enabled: true,
     id: 'qwen-max-latest',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 20,
@@ -56,6 +59,7 @@ const qwenChatModels: AIChatModelCard[] = [
       '通义千问超大规模语言模型，支持长文本上下文，以及基于长文档、多文档等多个场景的对话功能。',
     displayName: 'Qwen Long',
     id: 'qwen-long',
+    maxOutput: 6000,
     pricing: {
       currency: 'CNY',
       input: 0.5,
@@ -73,10 +77,11 @@ const qwenChatModels: AIChatModelCard[] = [
     displayName: 'Qwen VL Plus',
     enabled: true,
     id: 'qwen-vl-plus-latest',
+    maxOutput: 2048,
     pricing: {
       currency: 'CNY',
-      input: 8,
-      output: 8,
+      input: 1.5,
+      output: 4.5,
     },
     type: 'chat',
   },
@@ -84,16 +89,34 @@ const qwenChatModels: AIChatModelCard[] = [
     abilities: {
       vision: true,
     },
-    contextWindowTokens: 32_000,
+    contextWindowTokens: 32_768,
     description:
       '通义千问超大规模视觉语言模型。相比增强版，再次提升视觉推理能力和指令遵循能力，提供更高的视觉感知和认知水平。',
     displayName: 'Qwen VL Max',
     enabled: true,
     id: 'qwen-vl-max-latest',
+    maxOutput: 2048,
     pricing: {
       currency: 'CNY',
-      input: 20,
-      output: 20,
+      input: 3,
+      output: 9,
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      vision: true,
+    },
+    contextWindowTokens: 34_096,
+    description:
+      '通义千问OCR是文字提取专有模型，专注于文档、表格、试题、手写体文字等类型图像的文字提取能力。它能够识别多种文字，目前支持的语言有：汉语、英语、法语、日语、韩语、德语、俄语、意大利语、越南语、阿拉伯语。',
+    displayName: 'Qwen VL OCR',
+    id: 'qwen-vl-ocr-latest',
+    maxOutput: 4096,
+    pricing: {
+      currency: 'CNY',
+      input: 5,
+      output: 5,
     },
     type: 'chat',
   },
@@ -102,6 +125,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问数学模型是专门用于数学解题的语言模型。',
     displayName: 'Qwen Math Turbo',
     id: 'qwen-math-turbo-latest',
+    maxOutput: 3072,
     pricing: {
       currency: 'CNY',
       input: 2,
@@ -114,6 +138,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问数学模型是专门用于数学解题的语言模型。',
     displayName: 'Qwen Math Plus',
     id: 'qwen-math-plus-latest',
+    maxOutput: 3072,
     pricing: {
       currency: 'CNY',
       input: 4,
@@ -126,6 +151,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问代码模型。',
     displayName: 'Qwen Coder Turbo',
     id: 'qwen-coder-turbo-latest',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 2,
@@ -138,6 +164,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问代码模型。',
     displayName: 'Qwen Coder Plus',
     id: 'qwen-coder-plus-latest',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 3.5,
@@ -146,10 +173,14 @@ const qwenChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 32_768,
     description: 'QwQ模型是由 Qwen 团队开发的实验性研究模型，专注于增强 AI 推理能力。',
     displayName: 'QwQ 32B Preview',
     id: 'qwq-32b-preview',
+    maxOutput: 16_384,
     pricing: {
       currency: 'CNY',
       input: 3.5,
@@ -166,6 +197,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: 'QVQ模型是由 Qwen 团队开发的实验性研究模型，专注于提升视觉推理能力，尤其在数学推理领域。',
     displayName: 'QVQ 72B Preview',
     id: 'qvq-72b-preview',
+    maxOutput: 16_384,
     pricing: {
       currency: 'CNY',
       input: 12,
@@ -182,10 +214,11 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问2.5对外开源的7B规模的模型。',
     displayName: 'Qwen2.5 7B',
     id: 'qwen2.5-7b-instruct',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
-      input: 1,
-      output: 2,
+      input: 0.5,
+      output: 1,
     },
     type: 'chat',
   },
@@ -197,10 +230,11 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问2.5对外开源的14B规模的模型。',
     displayName: 'Qwen2.5 14B',
     id: 'qwen2.5-14b-instruct',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
-      input: 2,
-      output: 6,
+      input: 1,
+      output: 3,
     },
     type: 'chat',
   },
@@ -212,6 +246,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问2.5对外开源的32B规模的模型。',
     displayName: 'Qwen2.5 32B',
     id: 'qwen2.5-32b-instruct',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 3.5,
@@ -227,6 +262,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问2.5对外开源的72B规模的模型。',
     displayName: 'Qwen2.5 72B',
     id: 'qwen2.5-72b-instruct',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 4,
@@ -234,11 +270,29 @@ const qwenChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 1_000_000,
+    description: '通义千问2.5对外开源的72B规模的模型。',
+    displayName: 'Qwen2.5 14B 1M',
+    id: 'qwen2.5-14b-instruct-1m',
+    maxOutput: 8192,
+    pricing: {
+      currency: 'CNY',
+      input: 1,
+      output: 3,
+    },
+    releasedAt: '2025-01-27',
+    type: 'chat',
+  },
   {
     contextWindowTokens: 4096,
     description: 'Qwen-Math 模型具有强大的数学解题能力。',
     displayName: 'Qwen2.5 Math 7B',
     id: 'qwen2.5-math-7b-instruct',
+    maxOutput: 3072,
     pricing: {
       currency: 'CNY',
       input: 1,
@@ -251,6 +305,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: 'Qwen-Math 模型具有强大的数学解题能力。',
     displayName: 'Qwen2.5 Math 72B',
     id: 'qwen2.5-math-72b-instruct',
+    maxOutput: 3072,
     pricing: {
       currency: 'CNY',
       input: 4,
@@ -263,6 +318,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问代码模型开源版。',
     displayName: 'Qwen2.5 Coder 7B',
     id: 'qwen2.5-coder-7b-instruct',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 1,
@@ -275,6 +331,7 @@ const qwenChatModels: AIChatModelCard[] = [
     description: '通义千问代码模型开源版。',
     displayName: 'Qwen2.5 Coder 32B',
     id: 'qwen2.5-coder-32b-instruct',
+    maxOutput: 8192,
     pricing: {
       currency: 'CNY',
       input: 3.5,
@@ -312,6 +369,78 @@ const qwenChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
+  {
+    abilities: {
+      vision: true,
+    },
+    contextWindowTokens: 131_072,
+    description: 
+      '指令跟随、数学、解题、代码整体提升，万物识别能力提升，支持多样格式直接精准定位视觉元素，支持对长视频文件（最长10分钟）进行理解和秒级别的事件时刻定位，能理解时间先后和快慢，基于解析和定位能力支持操控OS或Mobile的Agent，关键信息抽取能力和Json格式输出能力强，此版本为72B版本，本系列能力最强的版本。',
+    displayName: 'Qwen2.5 VL 72B',
+    id: 'qwen2.5-vl-72b-instruct',
+    maxOutput: 2048,
+    pricing: {
+      currency: 'CNY',
+      input: 16,
+      output: 48,
+    },
+    releasedAt: '2025-01-27',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      vision: true,
+    },
+    contextWindowTokens: 131_072,
+    description: 
+      '指令跟随、数学、解题、代码整体提升，万物识别能力提升，支持多样格式直接精准定位视觉元素，支持对长视频文件（最长10分钟）进行理解和秒级别的事件时刻定位，能理解时间先后和快慢，基于解析和定位能力支持操控OS或Mobile的Agent，关键信息抽取能力和Json格式输出能力强，此版本为72B版本，本系列能力最强的版本。',
+    displayName: 'Qwen2.5 VL 7B',
+    id: 'qwen2.5-vl-7b-instruct',
+    maxOutput: 2048,
+    pricing: {
+      currency: 'CNY',
+      input: 2,
+      output: 5,
+    },
+    releasedAt: '2025-01-27',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      reasoning: true,
+    },
+    contextWindowTokens: 65_536,
+    description: 
+      'DeepSeek-R1 在后训练阶段大规模使用了强化学习技术，在仅有极少标注数据的情况下，极大提升了模型推理能力，尤其在数学、代码、自然语言推理等任务上。',
+    displayName: 'DeepSeek R1',
+    id: 'deepseek-r1',
+    maxOutput: 8192,
+    pricing: {
+      currency: 'CNY',
+      input: 0,
+      output: 0,
+    },
+    releasedAt: '2025-01-27',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 65_536,
+    description: 
+      'DeepSeek-V3 为自研 MoE 模型，671B 参数，激活 37B，在 14.8T token 上进行了预训练，在长文本、代码、数学、百科、中文能力上表现优秀。',
+    displayName: 'DeepSeek V3',
+    id: 'deepseek-v3',
+    maxOutput: 8192,
+    pricing: {
+      currency: 'CNY',
+      input: 0,
+      output: 0,
+    },
+    releasedAt: '2025-01-27',
+    type: 'chat',
+  },
 ];
 
 export const allModels = [...qwenChatModels];

package/src/config/modelProviders/github.ts

@@ -37,7 +37,7 @@ const Github: ModelProviderCard = {
       vision: true,
     },
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 134_144,
       description: '一种经济高效的AI解决方案，适用于多种文本和图像任务。',
       displayName: 'OpenAI GPT-4o mini',
       enabled: true,
@@ -47,15 +47,21 @@ const Github: ModelProviderCard = {
       vision: true,
     },
     {
-      contextWindowTokens: 128_000,
+      contextWindowTokens: 134_144,
       description: 'OpenAI GPT-4系列中最先进的多模态模型，可以处理文本和图像输入。',
       displayName: 'OpenAI GPT-4o',
       enabled: true,
       functionCall: true,
       id: 'gpt-4o',
-      maxOutput: 4096,
+      maxOutput: 16_384,
       vision: true,
     },
+    {
+      contextWindowTokens: 128_000,
+      displayName: 'DeepSeek R1',
+      id: 'DeepSeek-R1',
+      maxOutput: 4096,
+    },
     {
       contextWindowTokens: 262_144,
       description:
@@ -112,6 +118,12 @@ const Github: ModelProviderCard = {
       id: 'mistral-large',
       maxOutput: 4096,
     },
+    {
+      contextWindowTokens: 262_144,
+      displayName: 'Codestral',
+      id: 'Codestral-2501',
+      maxOutput: 4096,
+    },
     {
       contextWindowTokens: 131_072,
       description: '在高分辨率图像上表现出色的图像推理能力，适用于视觉理解应用。',
@@ -166,6 +178,18 @@ const Github: ModelProviderCard = {
       id: 'meta-llama-3-70b-instruct',
       maxOutput: 4096,
     },
+    {
+      contextWindowTokens: 16_384,
+      displayName: 'Phi 4',
+      id: 'Phi-4',
+      maxOutput: 16_384,
+    },
+    {
+      contextWindowTokens: 131_072,
+      displayName: 'Phi 3.5 MoE',
+      id: 'Phi-3.5-MoE-instruct',
+      maxOutput: 4096,
+    },
     {
       contextWindowTokens: 131_072,
       description: 'Phi-3-mini模型的更新版。',

package/src/config/modelProviders/qwen.ts

@@ -4,7 +4,7 @@ import { ModelProviderCard } from '@/types/llm';
 const Qwen: ModelProviderCard = {
   chatModels: [
     {
-      contextWindowTokens: 131_072,
+      contextWindowTokens: 1_000_000,
       description: '通义千问超大规模语言模型，支持中文、英文等不同语言输入。',
       displayName: 'Qwen Turbo',
       enabled: true,
@@ -64,13 +64,13 @@ const Qwen: ModelProviderCard = {
       id: 'qwen-vl-plus-latest',
       pricing: {
         currency: 'CNY',
-        input: 8,
-        output: 8,
+        input: 1.5,
+        output: 4.5,
       },
       vision: true,
     },
     {
-      contextWindowTokens: 32_000,
+      contextWindowTokens: 32_768,
       description:
         '通义千问超大规模视觉语言模型。相比增强版，再次提升视觉推理能力和指令遵循能力，提供更高的视觉感知和认知水平。',
       displayName: 'Qwen VL Max',
@@ -78,8 +78,21 @@ const Qwen: ModelProviderCard = {
       id: 'qwen-vl-max-latest',
       pricing: {
         currency: 'CNY',
-        input: 20,
-        output: 20,
+        input: 3,
+        output: 9,
+      },
+      vision: true,
+    },
+    {
+      contextWindowTokens: 34_096,
+      description:
+        '通义千问OCR是文字提取专有模型，专注于文档、表格、试题、手写体文字等类型图像的文字提取能力。它能够识别多种文字，目前支持的语言有：汉语、英语、法语、日语、韩语、德语、俄语、意大利语、越南语、阿拉伯语。',
+      displayName: 'Qwen VL OCR',
+      id: 'qwen-vl-ocr-latest',
+      pricing: {
+        currency: 'CNY',
+        input: 5,
+        output: 5,
       },
       vision: true,
     },
@@ -134,9 +147,22 @@ const Qwen: ModelProviderCard = {
       id: 'qwq-32b-preview',
       pricing: {
         currency: 'CNY',
-        input: 0,
-        output: 0,
+        input: 3.5,
+        output: 7,
+      },
+    },
+    {
+      contextWindowTokens: 32_768,
+      description: 'QVQ模型是由 Qwen 团队开发的实验性研究模型，专注于提升视觉推理能力，尤其在数学推理领域。',
+      displayName: 'QVQ 72B Preview',
+      id: 'qvq-72b-preview',
+      pricing: {
+        currency: 'CNY',
+        input: 12,
+        output: 36,
       },
+      releasedAt: '2024-12-25',
+      vision: true,
     },
     {
       contextWindowTokens: 131_072,
@@ -146,8 +172,8 @@ const Qwen: ModelProviderCard = {
       id: 'qwen2.5-7b-instruct',
       pricing: {
         currency: 'CNY',
-        input: 1,
-        output: 2,
+        input: 0.5,
+        output: 1,
       },
     },
     {
@@ -158,8 +184,8 @@ const Qwen: ModelProviderCard = {
       id: 'qwen2.5-14b-instruct',
       pricing: {
         currency: 'CNY',
-        input: 2,
-        output: 6,
+        input: 1,
+        output: 3,
       },
     },
     {
@@ -186,6 +212,18 @@ const Qwen: ModelProviderCard = {
         output: 12,
       },
     },
+    {
+      contextWindowTokens: 1_000_000,
+      description: '通义千问2.5对外开源的72B规模的模型。',
+      displayName: 'Qwen2.5 14B 1M',
+      functionCall: true,
+      id: 'qwen2.5-14b-instruct-1m',
+      pricing: {
+        currency: 'CNY',
+        input: 1,
+        output: 3,
+      },
+    },
     {
       contextWindowTokens: 4096,
       description: 'Qwen-Math 模型具有强大的数学解题能力。',
@@ -254,6 +292,46 @@ const Qwen: ModelProviderCard = {
       },
       vision: true,
     },
+    {
+      contextWindowTokens: 128_000,
+      description: 
+        '指令跟随、数学、解题、代码整体提升，万物识别能力提升，支持多样格式直接精准定位视觉元素，支持对长视频文件（最长10分钟）进行理解和秒级别的事件时刻定位，能理解时间先后和快慢，基于解析和定位能力支持操控OS或Mobile的Agent，关键信息抽取能力和Json格式输出能力强，此版本为72B版本，本系列能力最强的版本。',
+      displayName: 'Qwen2.5 VL 72B',
+      id: 'qwen2.5-vl-72b-instruct',
+      pricing: {
+        currency: 'CNY',
+        input: 16,
+        output: 48,
+      },
+      releasedAt: '2025-01-26',
+      vision: true,
+    },
+    {
+      contextWindowTokens: 65_536,
+      description: 
+        'DeepSeek-R1 在后训练阶段大规模使用了强化学习技术，在仅有极少标注数据的情况下，极大提升了模型推理能力，尤其在数学、代码、自然语言推理等任务上。',
+      displayName: 'DeepSeek R1',
+      id: 'deepseek-r1',
+      pricing: {
+        currency: 'CNY',
+        input: 0,
+        output: 0,
+      },
+      releasedAt: '2025-01-27',
+    },
+    {
+      contextWindowTokens: 65_536,
+      description: 
+        'DeepSeek-V3 为自研 MoE 模型，671B 参数，激活 37B，在 14.8T token 上进行了预训练，在长文本、代码、数学、百科、中文能力上表现优秀。',
+      displayName: 'DeepSeek V3',
+      id: 'deepseek-v3',
+      pricing: {
+        currency: 'CNY',
+        input: 0,
+        output: 0,
+      },
+      releasedAt: '2025-01-27',
+    },
   ],
   checkModel: 'qwen-turbo-latest',
   description:

package/src/database/repositories/dataImporter/index.ts

@@ -14,6 +14,7 @@ import {
 import { LobeChatDatabase } from '@/database/type';
 import { ImportResult } from '@/services/config';
 import { ImporterEntryData } from '@/types/importer';
+import { sanitizeUTF8 } from '@/utils/sanitizeUTF8';
 
 export class DataImporterRepos {
   private userId: string;
@@ -204,9 +205,10 @@ export class DataImporterRepos {
         // 2. insert messages
         if (shouldInsertMessages.length > 0) {
           const inertValues = shouldInsertMessages.map(
-            ({ id, extra, createdAt, updatedAt, sessionId, topicId, ...res }) => ({
+            ({ id, extra, createdAt, updatedAt, sessionId, topicId, content, ...res }) => ({
               ...res,
               clientId: id,
+              content: sanitizeUTF8(content),
               createdAt: new Date(createdAt),
               model: extra?.fromModel,
               parentId: null,

package/src/libs/langchain/loaders/pdf/index.ts

@@ -1,7 +1,7 @@
 import { PDFLoader } from '@langchain/community/document_loaders/fs/pdf';
 
 export const PdfLoader = async (fileBlob: Blob) => {
-  const loader = new PDFLoader(fileBlob);
+  const loader = new PDFLoader(fileBlob, { splitPages: true });
 
   return await loader.load();
 };

package/src/server/routers/async/file.ts

@@ -24,6 +24,7 @@ import {
   IAsyncTaskError,
 } from '@/types/asyncTask';
 import { safeParseJSON } from '@/utils/safeParseJSON';
+import { sanitizeUTF8 } from '@/utils/sanitizeUTF8';
 
 const fileProcedure = asyncAuthedProcedure.use(async (opts) => {
   const { ctx } = opts;
@@ -95,16 +96,13 @@ export const fileRouter = router({
                   ctx.jwtPayload,
                 );
 
-                const number = index + 1;
-                console.log(`执行第 ${number} 个任务`);
+                console.log(`run embedding task ${index + 1}`);
 
-                console.time(`任务[${number}]: embeddings`);
                 const embeddings = await agentRuntime.embeddings({
                   dimensions: 1024,
                   input: chunks.map((c) => c.text),
                   model,
                 });
-                console.timeEnd(`任务[${number}]: embeddings`);
 
                 const items: NewEmbeddingsItem[] =
                   embeddings?.map((e, idx) => ({
@@ -114,9 +112,7 @@ export const fileRouter = router({
                     model,
                   })) || [];
 
-                console.time(`任务[${number}]: insert db`);
                 await ctx.embeddingModel.bulkCreate(items);
-                console.timeEnd(`任务[${number}]: insert db`);
               },
               { concurrency: CONCURRENCY },
             );
@@ -215,7 +211,11 @@ export const fileRouter = router({
 
           // after finish partition, we need to filter out some elements
           const chunks = chunkResult.chunks.map(
-            (item): NewChunkItem => ({ ...item, userId: ctx.userId }),
+            ({ text, ...item }): NewChunkItem => ({
+              ...item,
+              text: text ? sanitizeUTF8(text) : '',
+              userId: ctx.userId,
+            }),
           );
 
           const duration = Date.now() - startAt;

package/src/utils/sanitizeUTF8.test.ts

@@ -0,0 +1,23 @@
+import { sanitizeUTF8 } from './sanitizeUTF8';
+
+describe('UTF-8 Sanitization', () => {
+  it('should handle null bytes', () => {
+    const input = 'test\u0000string';
+    expect(sanitizeUTF8(input)).toBe('teststring');
+  });
+
+  it('should handle invalid UTF-8 sequences', () => {
+    const input = 'test\uD800string'; // 未配对的代理项
+    expect(sanitizeUTF8(input)).toBe('teststring');
+  });
+
+  it('should handle invalid UTF-8 content', () => {
+    const input = '\u0002\u0000\u0000\u0002�{\\"error\\":{\\"code\\":\\"resource_exhausted\\",';
+    expect(sanitizeUTF8(input)).toBe('{\\"error\\":{\\"code\\":\\"resource_exhausted\\",');
+  });
+
+  it('should preserve valid UTF-8 characters', () => {
+    const input = '你好，世界！';
+    expect(sanitizeUTF8(input)).toBe('你好，世界！');
+  });
+});

package/src/utils/sanitizeUTF8.ts

@@ -0,0 +1,14 @@
+/**
+ * Sanitize UTF-8 string to remove all control characters and invalid code points.
+ * @param str
+ */
+export const sanitizeUTF8 = (str: string) => {
+  // 移除替换字符 (0xFFFD) 和其他非法字符
+  return (
+    str
+      .replaceAll('�', '') // 移除 Unicode 替换字符
+      // eslint-disable-next-line no-control-regex
+      .replaceAll(/[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F]/g, '') // 移除控制字符
+      .replaceAll(/[\uD800-\uDFFF]/g, '')
+  ); // 移除未配对的代理项码点
+};