@lobehub/chat 1.106.2 → 1.106.3

package/CHANGELOG.md

@@ -2,6 +2,40 @@
 
 # Changelog
 
+### [Version 1.106.3](https://github.com/lobehub/lobe-chat/compare/v1.106.2...v1.106.3)
+
+<sup>Released on **2025-07-29**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Moonshot assistant messages must not be empty.
+
+#### 💄 Styles
+
+- **misc**: Add volcengine kimi-k2 model, Add Zhipu GLM-4.5 models.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Moonshot assistant messages must not be empty, closes [#8419](https://github.com/lobehub/lobe-chat/issues/8419) ([a796495](https://github.com/lobehub/lobe-chat/commit/a796495))
+
+#### Styles
+
+- **misc**: Add volcengine kimi-k2 model, closes [#8591](https://github.com/lobehub/lobe-chat/issues/8591) ([9630167](https://github.com/lobehub/lobe-chat/commit/9630167))
+- **misc**: Add Zhipu GLM-4.5 models, closes [#8590](https://github.com/lobehub/lobe-chat/issues/8590) ([4f4620c](https://github.com/lobehub/lobe-chat/commit/4f4620c))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.106.2](https://github.com/lobehub/lobe-chat/compare/v1.106.1...v1.106.2)
 
 <sup>Released on **2025-07-29**</sup>

package/changelog/v1.json

@@ -1,4 +1,16 @@
 [
+  {
+    "children": {
+      "fixes": [
+        "Moonshot assistant messages must not be empty."
+      ],
+      "improvements": [
+        "Add volcengine kimi-k2 model, Add Zhipu GLM-4.5 models."
+      ]
+    },
+    "date": "2025-07-29",
+    "version": "1.106.3"
+  },
   {
     "children": {
       "fixes": [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.106.2",
+  "version": "1.106.3",
   "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/src/config/aiModels/google.ts

@@ -131,53 +131,6 @@ const googleChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
-  {
-    abilities: {
-      functionCall: true,
-      reasoning: true,
-      search: true,
-      vision: true,
-    },
-    contextWindowTokens: 1_048_576 + 65_536,
-    description: 'Gemini 2.5 Flash Preview 是 Google 性价比最高的模型，提供全面的功能。',
-    displayName: 'Gemini 2.5 Flash Preview 04-17',
-    id: 'gemini-2.5-flash-preview-04-17',
-    maxOutput: 65_536,
-    pricing: {
-      cachedInput: 0.0375,
-      input: 0.15,
-      output: 3.5, // Thinking
-    },
-    releasedAt: '2025-04-17',
-    settings: {
-      extendParams: ['thinkingBudget'],
-      searchImpl: 'params',
-      searchProvider: 'google',
-    },
-    type: 'chat',
-  },
-  {
-    abilities: {
-      functionCall: true,
-      reasoning: true,
-      search: true,
-      vision: true,
-    },
-    contextWindowTokens: 1_048_576 + 65_536,
-    description: 'Gemini 2.5 Flash Preview 是 Google 性价比最高的模型，提供全面的功能。',
-    displayName: 'Gemini 2.5 Flash Preview 04-17 for cursor testing',
-    id: 'gemini-2.5-flash-preview-04-17-thinking',
-    maxOutput: 65_536,
-    pricing: {
-      input: 0.15,
-      output: 3.5,
-    },
-    settings: {
-      searchImpl: 'params',
-      searchProvider: 'google',
-    },
-    type: 'chat',
-  },
   {
     abilities: {
       functionCall: true,
@@ -188,7 +141,6 @@ const googleChatModels: AIChatModelCard[] = [
     contextWindowTokens: 1_048_576 + 65_536,
     description: 'Gemini 2.5 Flash-Lite 是 Google 最小、性价比最高的模型，专为大规模使用而设计。',
     displayName: 'Gemini 2.5 Flash-Lite',
-    enabled: true,
     id: 'gemini-2.5-flash-lite',
     maxOutput: 65_536,
     pricing: {

package/src/config/aiModels/groq.ts

@@ -33,6 +33,7 @@ const groqChatModels: AIChatModelCard[] = [
     displayName: 'Kimi K2 Instruct',
     enabled: true,
     id: 'moonshotai/kimi-k2-instruct',
+    maxOutput: 16_384,
     pricing: {
       input: 1,
       output: 3,
@@ -53,6 +54,9 @@ const groqChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 131_072,
     displayName: 'Llama 4 Maverick (17Bx128E)',
     enabled: true,

package/src/config/aiModels/hunyuan.ts

@@ -44,6 +44,28 @@ const hunyuanChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
+  {
+    abilities: {
+      reasoning: true,
+      search: true,
+    },
+    contextWindowTokens: 92_000,
+    description:
+      '大幅提升高难度数学、逻辑和代码能力，优化模型输出稳定性，提升模型长文能力。',
+    displayName: 'Hunyuan T1 20250711',
+    id: 'hunyuan-t1-20250711',
+    maxOutput: 64_000,
+    pricing: {
+      currency: 'CNY',
+      input: 1,
+      output: 4,
+    },
+    releasedAt: '2025-07-11',
+    settings: {
+      searchImpl: 'params',
+    },
+    type: 'chat',
+  },
   {
     abilities: {
       reasoning: true,

package/src/config/aiModels/moonshot.ts

@@ -5,7 +5,6 @@ const moonshotChatModels: AIChatModelCard[] = [
   {
     abilities: {
       functionCall: true,
-      search: true,
     },
     contextWindowTokens: 131_072,
     description:
@@ -20,15 +19,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       output: 16,
     },
     releasedAt: '2025-07-11',
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
       vision: true,
     },
     contextWindowTokens: 131_072,
@@ -44,9 +39,6 @@ const moonshotChatModels: AIChatModelCard[] = [
       output: 30,
     },
     releasedAt: '2025-02-17',
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
@@ -71,7 +63,6 @@ const moonshotChatModels: AIChatModelCard[] = [
   {
     abilities: {
       functionCall: true,
-      search: true,
     },
     contextWindowTokens: 131_072,
     description: 'Moonshot V1 Auto 可以根据当前上下文占用的 Tokens 数量来选择合适的模型',
@@ -82,15 +73,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       input: 10, // 128k 上下文时
       output: 30,
     },
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
     },
     contextWindowTokens: 8192,
     description:
@@ -102,15 +89,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       input: 2,
       output: 10,
     },
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
     },
     contextWindowTokens: 32_768,
     description:
@@ -122,15 +105,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       input: 5,
       output: 20,
     },
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
     },
     contextWindowTokens: 131_072,
     description:
@@ -142,15 +121,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       input: 10,
       output: 30,
     },
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
       vision: true,
     },
     contextWindowTokens: 8192,
@@ -164,15 +139,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       output: 10,
     },
     releasedAt: '2025-01-14',
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
       vision: true,
     },
     contextWindowTokens: 32_768,
@@ -186,15 +157,11 @@ const moonshotChatModels: AIChatModelCard[] = [
       output: 20,
     },
     releasedAt: '2025-01-14',
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
   {
     abilities: {
       functionCall: true,
-      search: true,
       vision: true,
     },
     contextWindowTokens: 131_072,
@@ -208,9 +175,6 @@ const moonshotChatModels: AIChatModelCard[] = [
       output: 30,
     },
     releasedAt: '2025-01-14',
-    settings: {
-      searchImpl: 'params',
-    },
     type: 'chat',
   },
 ];

package/src/config/aiModels/qwen.ts

@@ -3,6 +3,89 @@ import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 // https://help.aliyun.com/zh/model-studio/models?spm=a2c4g.11186623
 
 const qwenChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 131_072,
+    description:
+      '总参数 1T，激活参数 32B。 非思维模型中，在前沿知识、数学和编码方面达到了顶尖水平，更擅长通用 Agent 任务。 针对代理任务进行了精心优化，不仅能回答问题，还能采取行动。 最适用于即兴、通用聊天和代理体验，是一款无需长时间思考的反射级模型。',
+    displayName: 'Kimi K2 Instruct',
+    enabled: true,
+    id: 'Moonshot-Kimi-K2-Instruct',
+    maxOutput: 8192,
+    organization: 'Qwen',
+    pricing: {
+      currency: 'CNY',
+      input: 4,
+      output: 16,
+    },
+    releasedAt: '2025-07-17',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    config: {
+      deploymentName: 'qwen3-coder-plus',
+    },
+    contextWindowTokens: 1_048_576,
+    description: '通义千问代码模型。最新的 Qwen3-Coder-Plus 系列模型是基于 Qwen3 的代码生成模型，具有强大的Coding Agent能力，擅长工具调用和环境交互，能够实现自主编程，代码能力卓越的同时兼具通用能力。',
+    displayName: 'Qwen3 Coder Plus',
+    id: 'qwen3-coder-plus',
+    maxOutput: 65_536,
+    organization: 'Qwen',
+    pricing: {
+      cachedInput: 2.4, // tokens 32K ~ 128K
+      currency: 'CNY',  
+      input: 6,
+      output: 24,
+    },
+    releasedAt: '2025-07-23',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+    },
+    contextWindowTokens: 131_072,
+    description:
+      '基于Qwen3的思考模式开源模型，相较上一版本（通义千问3-235B-A22B）逻辑能力、通用能力、知识增强及创作能力均有大幅提升，适用于高难度强推理场景。',
+    displayName: 'Qwen3 235B A22B Thinking 2507',
+    enabled: true,
+    id: 'qwen3-235b-a22b-thinking-2507',
+    maxOutput: 32_768,
+    organization: 'Qwen',
+    pricing: {
+      currency: 'CNY',
+      input: 2,
+      output: 20,
+    },
+    releasedAt: '2025-07-25',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 131_072,
+    description:
+      '基于Qwen3的非思考模式开源模型，相较上一版本（通义千问3-235B-A22B）主观创作能力与模型安全性均有小幅度提升。',
+    displayName: 'Qwen3 235B A22B Instruct 2507',
+    enabled: true,
+    id: 'qwen3-235b-a22b-instruct-2507',
+    maxOutput: 32_768,
+    organization: 'Qwen',
+    pricing: {
+      currency: 'CNY',
+      input: 2,
+      output: 8,
+    },
+    releasedAt: '2025-07-22',
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,
@@ -12,7 +95,6 @@ const qwenChatModels: AIChatModelCard[] = [
     description:
       'Qwen3是一款能力大幅提升的新一代通义千问大模型，在推理、通用、Agent和多语言等多个核心能力上均达到业界领先水平，并支持思考模式切换。',
     displayName: 'Qwen3 235B A22B',
-    enabled: true,
     id: 'qwen3-235b-a22b',
     maxOutput: 8192,
     organization: 'Qwen',
@@ -59,7 +141,6 @@ const qwenChatModels: AIChatModelCard[] = [
     description:
       'Qwen3是一款能力大幅提升的新一代通义千问大模型，在推理、通用、Agent和多语言等多个核心能力上均达到业界领先水平，并支持思考模式切换。',
     displayName: 'Qwen3 30B A3B',
-    enabled: true,
     id: 'qwen3-30b-a3b',
     maxOutput: 8192,
     organization: 'Qwen',
@@ -223,21 +304,22 @@ const qwenChatModels: AIChatModelCard[] = [
       search: true,
     },
     config: {
-      deploymentName: 'qwen-turbo-2025-04-28', // expired on 2025-10-26
+      deploymentName: 'qwen-turbo-2025-07-15',
     },
-    contextWindowTokens: 1_000_000,
+    contextWindowTokens: 1_000_000, // Non-thinking mode
     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',
-    maxOutput: 8192,
+    maxOutput: 16_384,
     organization: 'Qwen',
     pricing: {
+      cachedInput: 0.12,
       currency: 'CNY',
       input: 0.3,
       output: 3, // Thinking mode pricing
     },
-    releasedAt: '2025-04-28',
+    releasedAt: '2025-07-15',
     settings: {
       extendParams: ['enableReasoning', 'reasoningBudgetToken'],
       searchImpl: 'params',
@@ -251,21 +333,22 @@ const qwenChatModels: AIChatModelCard[] = [
       search: true,
     },
     config: {
-      deploymentName: 'qwen-plus-2025-04-28', // expired on 2025-10-26
+      deploymentName: 'qwen-plus-2025-07-14',
     },
     contextWindowTokens: 131_072,
     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: 8192,
+    maxOutput: 16_384,
     organization: 'Qwen',
     pricing: {
+      cachedInput: 0.32,
       currency: 'CNY',
       input: 0.8,
       output: 8, // Thinking mode pricing
     },
-    releasedAt: '2025-04-28',
+    releasedAt: '2025-07-14',
     settings: {
       extendParams: ['enableReasoning', 'reasoningBudgetToken'],
       searchImpl: 'params',
@@ -289,6 +372,7 @@ const qwenChatModels: AIChatModelCard[] = [
     maxOutput: 8192,
     organization: 'Qwen',
     pricing: {
+      cachedInput: 0.96,
       currency: 'CNY',
       input: 2.4,
       output: 9.6,
@@ -330,7 +414,6 @@ const qwenChatModels: AIChatModelCard[] = [
     description:
       'Qwen-Omni 系列模型支持输入多种模态的数据，包括视频、音频、图片、文本，并输出音频与文本。',
     displayName: 'Qwen Omni Turbo',
-    enabled: true,
     id: 'qwen-omni-turbo',
     maxOutput: 2048,
     organization: 'Qwen',
@@ -374,6 +457,7 @@ const qwenChatModels: AIChatModelCard[] = [
     maxOutput: 8192,
     organization: 'Qwen',
     pricing: {
+      cachedInput: 0.6,
       currency: 'CNY',
       input: 1.5,
       output: 4.5,
@@ -391,11 +475,11 @@ const qwenChatModels: AIChatModelCard[] = [
     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',
     maxOutput: 8192,
     organization: 'Qwen',
     pricing: {
+      cachedInput: 1.2,
       currency: 'CNY',
       input: 3,
       output: 9,
@@ -710,6 +794,21 @@ const qwenChatModels: AIChatModelCard[] = [
       input: 4,
       output: 12,
     },
+    releasedAt: '2025-07-23',
+    type: 'chat',
+  },
+  {
+    contextWindowTokens: 262_144,
+    description: '通义千问代码模型开源版。最新的 qwen3-coder-480b-a35b-instruct 是基于 Qwen3 的代码生成模型，具有强大的Coding Agent能力，擅长工具调用和环境交互，能够实现自主编程、代码能力卓越的同时兼具通用能力。',
+    displayName: 'Qwen3 Coder 480B A35B',
+    id: 'qwen3-coder-480b-a35b-instruct',
+    maxOutput: 65_536,
+    organization: 'Qwen',
+    pricing: {
+      currency: 'CNY',
+      input: 9, // tokens 32K ~ 128K
+      output: 36,
+    },
     type: 'chat',
   },
   {

package/src/config/aiModels/siliconcloud.ts

@@ -2,6 +2,42 @@ import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 // https://siliconflow.cn/zh-cn/models
 const siliconcloudChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+    },
+    contextWindowTokens: 131_072,
+    description:
+      'GLM-4.5 是一款专为智能体应用打造的基础模型，使用了混合专家（Mixture-of-Experts）架构。在工具调用、网页浏览、软件工程、前端编程领域进行了深度优化，支持无缝接入 Claude Code、Roo Code 等代码智能体中使用。GLM-4.5 采用混合推理模式，可以适应复杂推理和日常使用等多种应用场景。',
+    displayName: 'GLM-4.5',
+    id: 'zai-org/GLM-4.5',
+    pricing: {
+      currency: 'CNY',
+      input: 3.5,
+      output: 14,
+    },
+    releasedAt: '2025-07-28',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+    },
+    contextWindowTokens: 131_072,
+    description:
+      'GLM-4.5-Air 是一款专为智能体应用打造的基础模型，使用了混合专家（Mixture-of-Experts）架构。在工具调用、网页浏览、软件工程、前端编程领域进行了深度优化，支持无缝接入 Claude Code、Roo Code 等代码智能体中使用。GLM-4.5 采用混合推理模式，可以适应复杂推理和日常使用等多种应用场景。',
+    displayName: 'GLM-4.5-Air',
+    id: 'zai-org/GLM-4.5-Air',
+    pricing: {
+      currency: 'CNY',
+      input: 1,
+      output: 6,
+    },
+    releasedAt: '2025-07-28',
+    type: 'chat',
+  },
   {
     abilities: {
       reasoning: true,
@@ -33,6 +69,34 @@ const siliconcloudChatModels: AIChatModelCard[] = [
     releasedAt: '2025-06-30',
     type: 'chat',
   },
+  {
+    contextWindowTokens: 131_072,
+    description:
+      'Kimi K2 是一款具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
+    displayName: 'Kimi K2 Instruct',
+    id: 'moonshotai/Kimi-K2-Instruct',
+    pricing: {
+      currency: 'CNY',
+      input: 4,
+      output: 16,
+    },
+    releasedAt: '2025-07-11',
+    type: 'chat',
+  },
+  {
+    contextWindowTokens: 131_072,
+    description:
+      'Kimi K2 是一款具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
+    displayName: 'Kimi K2 Instruct (Pro)',
+    id: 'Pro/moonshotai/Kimi-K2-Instruct',
+    pricing: {
+      currency: 'CNY',
+      input: 4,
+      output: 16,
+    },
+    releasedAt: '2025-07-11',
+    type: 'chat',
+  },
   {
     abilities: {
       reasoning: true,
@@ -101,6 +165,43 @@ const siliconcloudChatModels: AIChatModelCard[] = [
     releasedAt: '2025-05-26',
     type: 'chat',
   },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+    },
+    contextWindowTokens: 262_144,
+    description:
+      'Qwen3-235B-A22B-Thinking-2507 是由阿里巴巴通义千问团队开发的 Qwen3 系列大型语言模型中的一员，专注于高难度的复杂推理任务。该模型基于混合专家（MoE）架构，总参数量达 2350 亿，而在处理每个 token 时仅激活约 220 亿参数，从而在保持强大性能的同时提高了计算效率。作为一个专门的“思考”模型，它在逻辑推理、数学、科学、编程和学术基准测试等需要人类专业知识的任务上表现显著提升，达到了开源思考模型中的顶尖水平。此外，模型还增强了通用能力，如指令遵循、工具使用和文本生成，并原生支持 256K 的长上下文理解能力，非常适合用于需要深度推理和处理长文档的场景。',
+    displayName: 'Qwen3 235B A22B Thinking 2507',
+    id: 'Qwen/Qwen3-235B-A22B-Thinking-2507',
+    organization: 'Qwen',
+    pricing: {
+      currency: 'CNY',
+      input: 2.5,
+      output: 10,
+    },
+    releasedAt: '2025-07-25',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 262_144,
+    description:
+      'Qwen3-235B-A22B-Instruct-2507 是由阿里云通义千问团队开发的 Qwen3 系列中的一款旗舰级混合专家（MoE）大语言模型。该模型拥有 2350 亿总参数，每次推理激活 220 亿参数。它是作为 Qwen3-235B-A22B 非思考模式的更新版本发布的，专注于在指令遵循、逻辑推理、文本理解、数学、科学、编程及工具使用等通用能力上实现显著提升。此外，模型增强了对多语言长尾知识的覆盖，并能更好地对齐用户在主观和开放性任务上的偏好，以生成更有帮助和更高质量的文本。',
+    displayName: 'Qwen3 235B A22B Instruct 2507',
+    id: 'Qwen/Qwen3-235B-A22B-Instruct-2507',
+    organization: 'Qwen',
+    pricing: {
+      currency: 'CNY',
+      input: 2.5,
+      output: 10,
+    },
+    releasedAt: '2025-07-21',
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,

package/src/config/aiModels/stepfun.ts

@@ -26,25 +26,6 @@ const stepfunChatModels: AIChatModelCard[] = [
     // },
     type: 'chat',
   },
-  {
-    abilities: {
-      functionCall: true,
-      search: true,
-    },
-    contextWindowTokens: 8000,
-    description: '高速模型，适合实时对话。',
-    displayName: 'Step 1 Flash',
-    id: 'step-1-flash', // 将在2025年4月30日下线
-    pricing: {
-      currency: 'CNY',
-      input: 1,
-      output: 4,
-    },
-    settings: {
-      searchImpl: 'params',
-    },
-    type: 'chat',
-  },
   {
     abilities: {
       functionCall: true,
@@ -83,25 +64,6 @@ const stepfunChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
-  {
-    abilities: {
-      functionCall: true,
-      search: true,
-    },
-    contextWindowTokens: 128_000,
-    description: '平衡性能与成本，适合一般场景。',
-    displayName: 'Step 1 128K',
-    id: 'step-1-128k', // 将在2025年4月30日下线
-    pricing: {
-      currency: 'CNY',
-      input: 40,
-      output: 200,
-    },
-    settings: {
-      searchImpl: 'params',
-    },
-    type: 'chat',
-  },
   {
     abilities: {
       functionCall: true,
@@ -240,21 +202,6 @@ const stepfunChatModels: AIChatModelCard[] = [
     releasedAt: '2025-01-22',
     type: 'chat',
   },
-  {
-    abilities: {
-      vision: true,
-    },
-    contextWindowTokens: 32_000,
-    description: '该模型拥有强大的视频理解能力。',
-    displayName: 'Step 1.5V Mini',
-    id: 'step-1.5v-mini',
-    pricing: {
-      currency: 'CNY',
-      input: 8,
-      output: 35,
-    },
-    type: 'chat',
-  },
   {
     abilities: {
       vision: true,

package/src/config/aiModels/volcengine.ts

@@ -4,6 +4,27 @@ import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 // pricing https://console.volcengine.com/ark/region:ark+cn-beijing/openManagement
 
 const doubaoChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+    },
+    config: {
+      deploymentName: 'kimi-k2-250711',
+    },
+    contextWindowTokens: 256_000,
+    description:
+      'Kimi-K2 是一款Moonshot AI推出的具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
+    displayName: 'Kimi K2',
+    enabled: true,
+    id: 'kimi-k2',
+    maxOutput: 16_384,
+    pricing: {
+      currency: 'CNY',
+      input: 4,
+      output: 16,
+    },
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,

package/src/config/aiModels/zhipu.ts

@@ -1,17 +1,140 @@
 import { AIChatModelCard, AIImageModelCard } from '@/types/aiModel';
 
 const zhipuChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      search: true,
+    },
+    contextWindowTokens: 128_000,
+    description:
+      '智谱最新旗舰模型，支持思考模式切换，综合能力达到开源模型的 SOTA 水平，上下文长度可达128K。',
+    displayName: 'GLM-4.5',
+    enabled: true,
+    id: 'glm-4.5',
+    maxOutput: 32_768,
+    pricing: {
+      cachedInput: 0.8,
+      currency: 'CNY',
+      input: 4, // 输入长度 [32k, 128k]
+      output: 16,
+    },
+    settings: {
+      extendParams: ['enableReasoning'],
+      searchImpl: 'params',
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      search: true,
+    },
+    contextWindowTokens: 128_000,
+    description:
+      'GLM-4.5 的极速版，在性能强劲的同时，生成速度可达 100 tokens/秒。',
+    displayName: 'GLM-4.5-X',
+    id: 'glm-4.5-x',
+    maxOutput: 32_768,
+    pricing: {
+      cachedInput: 3.2,
+      currency: 'CNY',
+      input: 16, // 输入长度 [32k, 128k]
+      output: 64,
+    },
+    settings: {
+      extendParams: ['enableReasoning'],
+      searchImpl: 'params',
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      search: true,
+    },
+    contextWindowTokens: 128_000,
+    description:
+      'GLM-4.5 的轻量版，兼顾性能与性价比，可灵活切换混合思考模型。',
+    displayName: 'GLM-4.5-Air',
+    id: 'glm-4.5-air',
+    maxOutput: 32_768,
+    pricing: {
+      cachedInput: 0.24,
+      currency: 'CNY',
+      input: 1.2, // 输入长度 [32k, 128k]
+      output: 8,
+    },
+    settings: {
+      extendParams: ['enableReasoning'],
+      searchImpl: 'params',
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      search: true,
+    },
+    contextWindowTokens: 128_000,
+    description:
+      'GLM-4.5-Air 的极速版，响应速度更快，专为大规模高速度需求打造。',
+    displayName: 'GLM-4.5-AirX',
+    id: 'glm-4.5-airx',
+    maxOutput: 32_768,
+    pricing: {
+      cachedInput: 1.6,
+      currency: 'CNY',
+      input: 8, // 输入长度 [32k, 128k]
+      output: 32,
+    },
+    settings: {
+      extendParams: ['enableReasoning'],
+      searchImpl: 'params',
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      search: true,
+    },
+    contextWindowTokens: 128_000,
+    description:
+      'GLM-4.5 的免费版，推理、代码、智能体等任务表现出色。',
+    displayName: 'GLM-4.5-Flash',
+    enabled: true,
+    id: 'glm-4.5-flash',
+    maxOutput: 32_768,
+    pricing: {
+      cachedInput: 0,
+      currency: 'CNY',
+      input: 0,
+      output: 0,
+    },
+    settings: {
+      extendParams: ['enableReasoning'],
+      searchImpl: 'params',
+    },
+    type: 'chat',
+  },
   {
     abilities: {
       reasoning: true,
       search: true,
       vision: true,
     },
-    contextWindowTokens: 64_000,
+    contextWindowTokens: 65_536,
     description:
       'GLM-4.1V-Thinking 系列模型是目前已知10B级别的VLM模型中性能最强的视觉模型，融合了同级别SOTA的各项视觉语言任务，包括视频理解、图片问答、学科解题、OCR文字识别、文档和图表解读、GUI Agent、前端网页Coding、Grounding等，多项任务能力甚至超过8倍参数量的Qwen2.5-VL-72B。通过领先的强化学习技术，模型掌握了通过思维链推理的方式提升回答的准确性和丰富度，从最终效果和可解释性等维度都显著超过传统的非thinking模型。',
     displayName: 'GLM-4.1V-Thinking-FlashX',
     id: 'glm-4.1v-thinking-flashx',
+    maxOutput: 16_384,
     pricing: {
       currency: 'CNY',
       input: 2,
@@ -28,12 +151,13 @@ const zhipuChatModels: AIChatModelCard[] = [
       search: true,
       vision: true,
     },
-    contextWindowTokens: 64_000,
+    contextWindowTokens: 65_536,
     description:
       'GLM-4.1V-Thinking 系列模型是目前已知10B级别的VLM模型中性能最强的视觉模型，融合了同级别SOTA的各项视觉语言任务，包括视频理解、图片问答、学科解题、OCR文字识别、文档和图表解读、GUI Agent、前端网页Coding、Grounding等，多项任务能力甚至超过8倍参数量的Qwen2.5-VL-72B。通过领先的强化学习技术，模型掌握了通过思维链推理的方式提升回答的准确性和丰富度，从最终效果和可解释性等维度都显著超过传统的非thinking模型。',
     displayName: 'GLM-4.1V-Thinking-Flash',
     enabled: true,
     id: 'glm-4.1v-thinking-flash',
+    maxOutput: 16_384,
     pricing: {
       currency: 'CNY',
       input: 0,
@@ -68,7 +192,7 @@ const zhipuChatModels: AIChatModelCard[] = [
     description: '推理模型: 具备强大推理能力，适用于需要深度推理的任务。',
     displayName: 'GLM-Z1-Air',
     id: 'glm-z1-air',
-    maxOutput: 32_000,
+    maxOutput: 32_768,
     pricing: {
       currency: 'CNY',
       input: 0.5,
@@ -84,7 +208,7 @@ const zhipuChatModels: AIChatModelCard[] = [
       reasoning: true,
       search: true,
     },
-    contextWindowTokens: 32_000,
+    contextWindowTokens: 32_768,
     description: '极速推理：具有超快的推理速度和强大的推理效果。',
     displayName: 'GLM-Z1-AirX',
     id: 'glm-z1-airx',
@@ -108,7 +232,7 @@ const zhipuChatModels: AIChatModelCard[] = [
     description: '高速低价：Flash增强版本，超快推理速度，更快并发保障。',
     displayName: 'GLM-Z1-FlashX',
     id: 'glm-z1-flashx',
-    maxOutput: 32_000,
+    maxOutput: 32_768,
     pricing: {
       currency: 'CNY',
       input: 0.1,
@@ -126,11 +250,10 @@ const zhipuChatModels: AIChatModelCard[] = [
     },
     contextWindowTokens: 128_000,
     description:
-      'GLM-Z1 系列具备强大的复杂推理能力，在逻辑推理、数学、编程等领域表现优异。最大上下文长度为32K。',
+      'GLM-Z1 系列具备强大的复杂推理能力，在逻辑推理、数学、编程等领域表现优异。',
     displayName: 'GLM-Z1-Flash',
-    enabled: true,
     id: 'glm-z1-flash',
-    maxOutput: 32_000,
+    maxOutput: 32_768,
     pricing: {
       currency: 'CNY',
       input: 0,
@@ -149,7 +272,6 @@ const zhipuChatModels: AIChatModelCard[] = [
     contextWindowTokens: 128_000,
     description: 'GLM-4-Flash 是处理简单任务的理想选择，速度最快且免费。',
     displayName: 'GLM-4-Flash-250414',
-    enabled: true,
     id: 'glm-4-flash-250414',
     maxOutput: 16_000,
     pricing: {
@@ -328,7 +450,6 @@ const zhipuChatModels: AIChatModelCard[] = [
     description:
       'GLM-4V-Flash 专注于高效的单一图像理解，适用于快速图像解析的场景，例如实时图像分析或批量图像处理。',
     displayName: 'GLM-4V-Flash',
-    enabled: true,
     id: 'glm-4v-flash',
     pricing: {
       currency: 'CNY',
@@ -374,7 +495,7 @@ const zhipuChatModels: AIChatModelCard[] = [
       'CodeGeeX-4 是强大的AI编程助手，支持多种编程语言的智能问答与代码补全，提升开发效率。',
     displayName: 'CodeGeeX-4',
     id: 'codegeex-4',
-    maxOutput: 32_000,
+    maxOutput: 32_768,
     pricing: {
       currency: 'CNY',
       input: 0.1,

package/src/config/modelProviders/moonshot.ts

@@ -39,6 +39,7 @@ const Moonshot: ModelProviderCard = {
   modelsUrl: 'https://platform.moonshot.cn/docs/intro',
   name: 'Moonshot',
   settings: {
+    disableBrowserRequest: true, // CORS error
     proxyUrl: {
       placeholder: 'https://api.moonshot.cn/v1',
     },

package/src/libs/model-runtime/moonshot/index.ts

@@ -11,7 +11,15 @@ export const LobeMoonshotAI = createOpenAICompatibleRuntime({
   baseURL: 'https://api.moonshot.cn/v1',
   chatCompletion: {
     handlePayload: (payload: ChatStreamPayload) => {
-      const { enabledSearch, temperature, tools, ...rest } = payload;
+      const { enabledSearch, messages, temperature, tools, ...rest } = payload;
+
+      // 为 assistant 空消息添加一个空格 (#8418)
+      const filteredMessages = messages.map(message => {
+        if (message.role === 'assistant' && (!message.content || message.content === '')) {
+          return { ...message, content: ' ' };
+        }
+        return message;
+      });
 
       const moonshotTools = enabledSearch
         ? [
@@ -27,6 +35,7 @@ export const LobeMoonshotAI = createOpenAICompatibleRuntime({
 
       return {
         ...rest,
+        messages: filteredMessages,
         temperature: temperature !== undefined ? temperature / 2 : undefined,
         tools: moonshotTools,
       } as any;

package/src/libs/model-runtime/utils/modelParse.ts

@@ -7,7 +7,7 @@ export interface ModelProcessorConfig {
   visionKeywords?: readonly string[];
 }
 
-// 模型标签关键词配置
+// 模型能力标签关键词配置
 export const MODEL_LIST_CONFIGS = {
   anthropic: {
     functionCallKeywords: ['claude'],
@@ -64,7 +64,7 @@ export const MODEL_LIST_CONFIGS = {
   },
   zhipu: {
     functionCallKeywords: ['glm-4', 'glm-z1'],
-    reasoningKeywords: ['glm-zero', 'glm-z1'],
+    reasoningKeywords: ['glm-zero', 'glm-z1', 'glm-4.5'],
     visionKeywords: ['glm-4v'],
   },
 } as const;

package/src/libs/model-runtime/zhipu/index.ts

@@ -1,6 +1,8 @@
 import { ModelProvider } from '../types';
 import { MODEL_LIST_CONFIGS, processModelList } from '../utils/modelParse';
 import { createOpenAICompatibleRuntime } from '../utils/openaiCompatibleFactory';
+import { OpenAIStream } from '../utils/streams/openai';
+import { convertIterableToStream } from '../utils/streams/protocol';
 
 export interface ZhipuModelCard {
   description: string;
@@ -12,7 +14,8 @@ export const LobeZhipuAI = createOpenAICompatibleRuntime({
   baseURL: 'https://open.bigmodel.cn/api/paas/v4',
   chatCompletion: {
     handlePayload: (payload) => {
-      const { enabledSearch, max_tokens, model, temperature, tools, top_p, ...rest } = payload;
+      const { enabledSearch, max_tokens, model, temperature, thinking, tools, top_p, ...rest } =
+        payload;
 
       const zhipuTools = enabledSearch
         ? [
@@ -39,6 +42,7 @@ export const LobeZhipuAI = createOpenAICompatibleRuntime({
               max_tokens,
         model,
         stream: true,
+        thinking: model.includes('-4.5') ? { type: thinking?.type } : undefined,
         tools: zhipuTools,
         ...(model === 'glm-4-alltools'
           ? {
@@ -54,6 +58,58 @@ export const LobeZhipuAI = createOpenAICompatibleRuntime({
             }),
       } as any;
     },
+    handleStream: (stream, { callbacks, inputStartAt }) => {
+      const readableStream =
+        stream instanceof ReadableStream ? stream : convertIterableToStream(stream);
+
+      // GLM-4.5 系列模型在 tool_calls 中返回的 index 为 -1，需要在进入 OpenAIStream 之前修正
+      // 因为 OpenAIStream 内部会过滤掉 index < 0 的 tool_calls (openai.ts:58-60)
+      const preprocessedStream = readableStream.pipeThrough(
+        new TransformStream({
+          transform(chunk, controller) {
+            // 处理原始的 OpenAI ChatCompletionChunk 格式
+            if (chunk.choices && chunk.choices[0]) {
+              const choice = chunk.choices[0];
+              if (choice.delta?.tool_calls && Array.isArray(choice.delta.tool_calls)) {
+                // 修正负数 index，将 -1 转换为基于数组位置的正数 index
+                const fixedToolCalls = choice.delta.tool_calls.map(
+                  (toolCall: any, globalIndex: number) => ({
+                    ...toolCall,
+                    index: toolCall.index < 0 ? globalIndex : toolCall.index,
+                  }),
+                );
+
+                // 创建修正后的 chunk
+                const fixedChunk = {
+                  ...chunk,
+                  choices: [
+                    {
+                      ...choice,
+                      delta: {
+                        ...choice.delta,
+                        tool_calls: fixedToolCalls,
+                      },
+                    },
+                  ],
+                };
+
+                controller.enqueue(fixedChunk);
+              } else {
+                controller.enqueue(chunk);
+              }
+            } else {
+              controller.enqueue(chunk);
+            }
+          },
+        }),
+      );
+
+      return OpenAIStream(preprocessedStream, {
+        callbacks,
+        inputStartAt,
+        provider: 'zhipu',
+      });
+    },
   },
   debug: {
     chatCompletion: () => process.env.DEBUG_ZHIPU_CHAT_COMPLETION === '1',