@lobehub/chat 1.70.0 → 1.70.1

package/CHANGELOG.md

@@ -2,6 +2,31 @@
 
 # Changelog
 
+### [Version 1.70.1](https://github.com/lobehub/lobe-chat/compare/v1.70.0...v1.70.1)
+
+<sup>Released on **2025-03-10**</sup>
+
+#### 🐛 Bug Fixes
+
+- **misc**: Fix anthropic max tokens.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### What's fixed
+
+- **misc**: Fix anthropic max tokens, closes [#6859](https://github.com/lobehub/lobe-chat/issues/6859) ([35fbc6c](https://github.com/lobehub/lobe-chat/commit/35fbc6c))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ## [Version 1.70.0](https://github.com/lobehub/lobe-chat/compare/v1.69.6...v1.70.0)
 
 <sup>Released on **2025-03-09**</sup>

package/changelog/v1.json

@@ -1,4 +1,13 @@
 [
+  {
+    "children": {
+      "fixes": [
+        "Fix anthropic max tokens."
+      ]
+    },
+    "date": "2025-03-10",
+    "version": "1.70.1"
+  },
   {
     "children": {
       "features": [

package/locales/ar/chat.json

@@ -65,7 +65,7 @@
     "warp": "تغيير السطر"
   },
   "intentUnderstanding": {
-    "title": "جارٍ تحليل وفهم نواياك..."
+    "title": "جارٍ فهم وتحليل نواياك..."
   },
   "knowledgeBase": {
     "all": "جميع المحتويات",

package/locales/bg-BG/chat.json

@@ -65,7 +65,7 @@
     "warp": "Нов ред"
   },
   "intentUnderstanding": {
-    "title": "Анализирам и разбирам вашето намерение..."
+    "title": "Разбирам и анализирам вашето намерение..."
   },
   "knowledgeBase": {
     "all": "Всички съдържания",

package/locales/de-DE/chat.json

@@ -65,7 +65,7 @@
     "warp": "Zeilenumbruch"
   },
   "intentUnderstanding": {
-    "title": "Analysiere und verstehe Ihre Absicht..."
+    "title": "Verstehe und analysiere gerade Ihre Absicht..."
   },
   "knowledgeBase": {
     "all": "Alle Inhalte",

package/locales/en-US/chat.json

@@ -65,7 +65,7 @@
     "warp": "New Line"
   },
   "intentUnderstanding": {
-    "title": "Analyzing and understanding your intent..."
+    "title": "Understanding and analyzing your intent..."
   },
   "knowledgeBase": {
     "all": "All Content",

package/locales/es-ES/chat.json

@@ -65,7 +65,7 @@
     "warp": "Salto de línea"
   },
   "intentUnderstanding": {
-    "title": "Analizando y comprendiendo su intención..."
+    "title": "Entendiendo y analizando su intención..."
   },
   "knowledgeBase": {
     "all": "Todo el contenido",

package/locales/fa-IR/chat.json

@@ -65,7 +65,7 @@
     "warp": "خط جدید"
   },
   "intentUnderstanding": {
-    "title": "در حال تحلیل و درک نیت شما..."
+    "title": "در حال درک و تحلیل نیت شما..."
   },
   "knowledgeBase": {
     "all": "همه محتوا",

package/locales/fr-FR/chat.json

@@ -65,7 +65,7 @@
     "warp": "Saut de ligne"
   },
   "intentUnderstanding": {
-    "title": "Analyse et comprend votre intention..."
+    "title": "En train de comprendre et d'analyser votre intention..."
   },
   "knowledgeBase": {
     "all": "Tout le contenu",

package/locales/it-IT/chat.json

@@ -65,7 +65,7 @@
     "warp": "A capo"
   },
   "intentUnderstanding": {
-    "title": "Analizzando e comprendendo le tue intenzioni..."
+    "title": "Stiamo comprendendo e analizzando la tua intenzione..."
   },
   "knowledgeBase": {
     "all": "Tutti i contenuti",

package/locales/ja-JP/chat.json

@@ -65,7 +65,7 @@
     "warp": "改行"
   },
   "intentUnderstanding": {
-    "title": "あなたの意図を分析し理解しています..."
+    "title": "あなたの意図を理解し、分析しています..."
   },
   "knowledgeBase": {
     "all": "すべてのコンテンツ",

package/locales/ko-KR/chat.json

@@ -65,7 +65,7 @@
     "warp": "줄바꿈"
   },
   "intentUnderstanding": {
-    "title": "귀하의 의도를 분석하고 이해하는 중입니다..."
+    "title": "귀하의 의도를 이해하고 분석하는 중입니다..."
   },
   "knowledgeBase": {
     "all": "모든 내용",

package/locales/nl-NL/chat.json

@@ -65,7 +65,7 @@
     "warp": "Nieuwe regel"
   },
   "intentUnderstanding": {
-    "title": "Bezig met het analyseren en begrijpen van uw intentie..."
+    "title": "Bezig met het begrijpen en analyseren van uw intentie..."
   },
   "knowledgeBase": {
     "all": "Alle inhoud",

package/locales/pl-PL/chat.json

@@ -65,7 +65,7 @@
     "warp": "Złamanie wiersza"
   },
   "intentUnderstanding": {
-    "title": "Analizuję i rozumiem Twoje intencje..."
+    "title": "Rozumiemy i analizujemy Twoje intencje..."
   },
   "knowledgeBase": {
     "all": "Wszystkie treści",

package/locales/pt-BR/chat.json

@@ -65,7 +65,7 @@
     "warp": "Quebrar linha"
   },
   "intentUnderstanding": {
-    "title": "Analisando e compreendendo sua intenção..."
+    "title": "Entendendo e analisando sua intenção..."
   },
   "knowledgeBase": {
     "all": "Todo conteúdo",

package/locales/ru-RU/chat.json

@@ -65,7 +65,7 @@
     "warp": "Перенос строки"
   },
   "intentUnderstanding": {
-    "title": "Анализ и понимание вашего намерения..."
+    "title": "Мы понимаем и анализируем ваше намерение..."
   },
   "knowledgeBase": {
     "all": "Все содержимое",

package/locales/tr-TR/chat.json

@@ -65,7 +65,7 @@
     "warp": "Satır atla"
   },
   "intentUnderstanding": {
-    "title": "Niyetinizi analiz ediyor ve anlıyor..."
+    "title": "Niyetinizi anlama ve analiz etme aşamasındayız..."
   },
   "knowledgeBase": {
     "all": "Tüm İçerik",

package/locales/vi-VN/chat.json

@@ -65,7 +65,7 @@
     "warp": "Xuống dòng"
   },
   "intentUnderstanding": {
-    "title": "Đang phân tích và hiểu ý định của bạn..."
+    "title": "Đang hiểu và phân tích ý định của bạn..."
   },
   "knowledgeBase": {
     "all": "Tất cả nội dung",

package/locales/zh-CN/chat.json

@@ -65,7 +65,7 @@
     "warp": "换行"
   },
   "intentUnderstanding": {
-    "title": "正在分析并理解意图您的意图..."
+    "title": "正在理解并分析您的意图..."
   },
   "knowledgeBase": {
     "all": "所有内容",

package/locales/zh-TW/chat.json

@@ -65,7 +65,7 @@
     "warp": "換行"
   },
   "intentUnderstanding": {
-    "title": "正在分析並理解您的意圖..."
+    "title": "正在理解並分析您的意圖..."
   },
   "knowledgeBase": {
     "all": "所有內容",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.70.0",
+  "version": "1.70.1",
   "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/libs/agent-runtime/anthropic/index.test.ts

@@ -1,7 +1,7 @@
 // @vitest-environment node
 import { Mock, afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
 
-import { ChatCompletionTool } from '@/libs/agent-runtime';
+import { ChatCompletionTool, ChatStreamPayload } from '@/libs/agent-runtime';
 
 import * as anthropicHelpers from '../utils/anthropicHelpers';
 import * as debugStreamModule from '../utils/debugStream';
@@ -511,5 +511,319 @@ describe('LobeAnthropicAI', () => {
         ).resolves.toBeInstanceOf(Response);
       });
     });
+
+    describe('buildAnthropicPayload', () => {
+      it('should correctly build payload with user messages only', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.5,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 4096,
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.25,
+        });
+      });
+
+      it('should correctly build payload with system message', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [
+            { content: 'You are a helpful assistant', role: 'system' },
+            { content: 'Hello', role: 'user' },
+          ],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 4096,
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          system: [
+            {
+              cache_control: { type: 'ephemeral' },
+              text: 'You are a helpful assistant',
+              type: 'text',
+            },
+          ],
+          temperature: 0.35,
+        });
+      });
+
+      it('should correctly build payload with tools', async () => {
+        const tools: ChatCompletionTool[] = [
+          { function: { name: 'tool1', description: 'desc1' }, type: 'function' },
+        ];
+
+        const spyOn = vi.spyOn(anthropicHelpers, 'buildAnthropicTools').mockReturnValueOnce([
+          {
+            name: 'tool1',
+            description: 'desc1',
+          },
+        ] as any);
+
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Use a tool', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.8,
+          tools,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 4096,
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Use a tool', type: 'text' }],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.4,
+          tools: [{ name: 'tool1', description: 'desc1' }],
+        });
+
+        expect(spyOn).toHaveBeenCalledWith(tools, {
+          enabledContextCaching: true,
+        });
+      });
+
+      it('should correctly build payload with thinking mode enabled', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Solve this problem', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.9,
+          thinking: { type: 'enabled', budget_tokens: 0 },
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 64000,
+          messages: [
+            {
+              content: [
+                { cache_control: { type: 'ephemeral' }, text: 'Solve this problem', type: 'text' },
+              ],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          thinking: { type: 'enabled', budget_tokens: 0 },
+        });
+      });
+
+      it('should respect max_tokens in thinking mode when provided', async () => {
+        const payload: ChatStreamPayload = {
+          max_tokens: 1000,
+          messages: [{ content: 'Solve this problem', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+          thinking: { type: 'enabled', budget_tokens: 0 },
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 1000,
+          messages: [
+            {
+              content: [
+                { cache_control: { type: 'ephemeral' }, text: 'Solve this problem', type: 'text' },
+              ],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          thinking: { type: 'enabled', budget_tokens: 0 },
+        });
+      });
+
+      it('should use budget_tokens in thinking mode when provided', async () => {
+        const payload: ChatStreamPayload = {
+          max_tokens: 1000,
+          messages: [{ content: 'Solve this problem', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.5,
+          thinking: { type: 'enabled', budget_tokens: 2000 },
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 3000, // budget_tokens + max_tokens
+          messages: [
+            {
+              content: [
+                { cache_control: { type: 'ephemeral' }, text: 'Solve this problem', type: 'text' },
+              ],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          thinking: { type: 'enabled', budget_tokens: 2000 },
+        });
+      });
+
+      it('should cap max_tokens at 64000 in thinking mode', async () => {
+        const payload: ChatStreamPayload = {
+          max_tokens: 10000,
+          messages: [{ content: 'Solve this problem', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.6,
+          thinking: { type: 'enabled', budget_tokens: 60000 },
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result).toEqual({
+          max_tokens: 64000, // capped at 64000
+          messages: [
+            {
+              content: [
+                { cache_control: { type: 'ephemeral' }, text: 'Solve this problem', type: 'text' },
+              ],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          thinking: { type: 'enabled', budget_tokens: 60000 },
+        });
+      });
+
+      it('should set correct max_tokens based on model for claude-3 models', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result.max_tokens).toBe(4096);
+      });
+
+      it('should set correct max_tokens based on model for non claude-3 models', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-2.1',
+          temperature: 0.7,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result.max_tokens).toBe(8192);
+      });
+
+      it('should respect max_tokens when explicitly provided', async () => {
+        const payload: ChatStreamPayload = {
+          max_tokens: 2000,
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result.max_tokens).toBe(2000);
+      });
+
+      it('should correctly handle temperature scaling', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 1.0,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result.temperature).toBe(0.5); // Anthropic uses 0-1 scale, so divide by 2
+      });
+
+      it('should not include temperature when not provided in payload', async () => {
+        // We need to create a partial payload without temperature
+        // but since the type requires it, we'll use type assertion
+        const partialPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+        } as ChatStreamPayload;
+
+        // Delete the temperature property to simulate it not being provided
+        delete (partialPayload as any).temperature;
+
+        const result = await instance['buildAnthropicPayload'](partialPayload);
+
+        expect(result.temperature).toBeUndefined();
+      });
+
+      it('should not include top_p when thinking is enabled', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+          thinking: { type: 'enabled', budget_tokens: 0 },
+          top_p: 0.9,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result.top_p).toBeUndefined();
+      });
+
+      it('should include top_p when thinking is not enabled', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+          top_p: 0.9,
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        expect(result.top_p).toBe(0.9);
+      });
+
+      it('should handle thinking with type disabled', async () => {
+        const payload: ChatStreamPayload = {
+          messages: [{ content: 'Hello', role: 'user' }],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.7,
+          thinking: { type: 'disabled', budget_tokens: 0 },
+        };
+
+        const result = await instance['buildAnthropicPayload'](payload);
+
+        // When thinking is disabled, it should be treated as if thinking wasn't provided
+        expect(result).toEqual({
+          max_tokens: 4096,
+          messages: [
+            {
+              content: [{ cache_control: { type: 'ephemeral' }, text: 'Hello', type: 'text' }],
+              role: 'user',
+            },
+          ],
+          model: 'claude-3-haiku-20240307',
+          temperature: 0.35,
+        });
+      });
+    });
   });
 });

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

@@ -100,10 +100,13 @@ export class LobeAnthropicAI implements LobeRuntimeAI {
 
     const postTools = buildAnthropicTools(tools, { enabledContextCaching });
 
-    if (!!thinking) {
-      const maxTokens =
-        // claude 3.7 thinking has max output of 64000 tokens
-        max_tokens ?? (thinking?.budget_tokens ? thinking?.budget_tokens + 64_000 : 8192);
+    if (!!thinking && thinking.type === 'enabled') {
+      // claude 3.7 thinking has max output of 64000 tokens
+      const maxTokens = !!max_tokens
+        ? thinking?.budget_tokens && thinking?.budget_tokens > max_tokens
+          ? Math.min(thinking?.budget_tokens + max_tokens, 64_000)
+          : max_tokens
+        : 64_000;
 
       // `temperature` may only be set to 1 when thinking is enabled.
       // `top_p` must be unset when thinking is enabled.

package/src/locales/default/chat.ts

@@ -66,7 +66,7 @@ export default {
     warp: '换行',
   },
   intentUnderstanding: {
-    title: '正在分析并理解意图您的意图...',
+    title: '正在理解并分析您的意图...',
   },
   knowledgeBase: {
     all: '所有内容',