GEO
赞助商内容
AI大模型

如何零成本使用Claude Code？2026年免费LLM代理（NVIDIA NIM/DeepSeek）全攻略

2026/4/25
如何零成本使用Claude Code？2026年免费LLM代理（NVIDIA NIM/DeepSeek）全攻略

AI Summary (BLUF)

free-claude-code is an open-source project that enables free use of Claude Code by proxying requests to free or low-cost model services like NVIDIA NIM, while retaining Claude Code's full engineering

Introduction

If you are looking for a free way to use Claude Code, the open-source project free-claude-code is definitely worth your attention. It allows Claude Code to seamlessly connect with various free or low-cost LLM services such as NVIDIA NIM, OpenRouter, and DeepSeek, while retaining the native engineering capabilities of Claude Code.

如果你正在寻找免费使用 Claude Code 的方法，那么 free-claude-code 这个开源项目绝对值得关注。它能让 Claude Code 完美对接 NVIDIA NIM、OpenRouter、DeepSeek 等多种免费或低成本的大模型服务，同时保留 Claude Code 原生的工程能力。

Why Choose free-claude-code?

As a Claude Code user, we all know its powerful engineering capabilities: context management, memory system, skill encapsulation, tool calling, and subagent control. However, the official API can be expensive, and there are various restrictions when using it domestically.

作为 Claude Code 用户，我们都知道它强大的工程能力：上下文管理、记忆系统、Skill 封装、工具调用、Subagent 控制等。但官方 API 价格不菲，国内使用还有各种限制。

free-claude-code perfectly addresses this issue.

free-claude-code 完美解决了这个问题。

Project Repository: https://github.com/Alishahryar1/free-claude-code

Core Advantages

核心优势

  1. Zero Cost: Directly calls the free NVIDIA NIM interface, with 40 requests per minute, which is more than sufficient.

    零成本使用：直接调用 NVIDIA NIM 免费接口，每分钟 40 次请求，完全够用。

  2. Perfect Disguise: Claude Code is completely unaware, believing it is still communicating with Anthropic's official servers.

    完美伪装：Claude Code 完全无感，以为还在和 Anthropic 官方通信。

  3. Chain-of-Thought Preservation: The thinking process of models like GLM and DeepSeek R1 is displayed perfectly.

    思考链保留：GLM、DeepSeek R1 等模型的思考过程完美显示。

  4. Tool Calling Restoration: For non-standard Tool Call outputs from open-source models, the proxy automatically parses them into a format that Claude can recognize.

    工具调用救活：开源模型不规范的 Tool Call 输出，代理自动解析成 Claude 能识别的格式。

  5. Intelligent Rate Limiting: Built-in concurrency control and 429 backoff mechanisms prevent rate limiting on free APIs.

    智能限流：内置并发控制和 429 退避，避免免费 API 被限流。

  6. Multi-Provider Mix-and-Match: A single configuration can simultaneously use models from different providers.

    多 Provider 混搭：一个配置同时使用不同提供商的模型。

Supported Model Providers

支持的模型提供商

中文版本 (Chinese Version)

提供商 特点 推荐模型
NVIDIA NIM 免费 40 次/分钟，注册即用 GLM-5.1、MiniMax M2.7、Kimi K2
OpenRouter 几百个模型，含免费款 deepseek-r1、gpt-oss-120b
DeepSeek 价格低，稳定 deepseek-reasoner、deepseek-chat
LM Studio 纯本地，数据不出网 本地 GGUF 模型
llama.cpp 完全本地运行 本地模型

English Version

Provider Characteristics Recommended Models
NVIDIA NIM Free 40 requests/min, ready upon registration GLM-5.1, MiniMax M2.7, Kimi K2
OpenRouter Hundreds of models, including free ones deepseek-r1, gpt-oss-120b
DeepSeek Low cost, stable deepseek-reasoner, deepseek-chat
LM Studio Purely local, data never leaves network Local GGUF models
llama.cpp Fully local operation Local models

Connecting to NVIDIA API: A Step-by-Step Guide

接入英伟达API流程

Step 1: Prepare Your NVIDIA API Key

第一步：准备 NVIDIA API Key

If you don't have one, go to build.nvidia.com to apply for a key starting with nvapi-.

如果没有，先去 build.nvidia.com 申请一个 nvapi- 开头的 Key。

Reference: NVIDIA Free API Application Process

参考：NVIDIA免费API的申请流程

Step 2: Clone the Project

第二步：克隆项目

git clone https://github.com/Alishahryar1/free-claude-code.git
cd free-claude-code

Step 3: Install Dependencies

第三步：安装依赖

This project uses Python and requires uv to be installed first:

这个项目使用 Python，需要先安装 uv：

pip install uv
uv sync

Step 4: Configure the .env File

第四步：配置 .env 文件

NVIDIA_NIM_API_KEY="nvapi-YourKey"

# Model mapping: all Claude models are mapped here
MODEL_OPUS="nvidia_nim/z-ai/glm-5.1"
MODEL_SONNET="nvidia_nim/z-ai/glm-5.1"
MODEL_HAIKU="nvidia_nim/z-ai/glm-5.1"
MODEL="nvidia_nim/z-ai/glm-5.1"

ENABLE_THINKING=true
PROVIDER_RATE_LIMIT=40

Step 5: Start the Proxy

第五步：启动代理

uv run uvicorn server:app --host 0.0.0.0 --port 8082

Step 6: Configure Claude Code

第六步：配置 Claude Code

Edit C:\Users\Administrator\.claude\settings.json:

编辑 C:\Users\Administrator\.claude\settings.json：

{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:8082",
    "ANTHROPIC_AUTH_TOKEN": "ccnim"
  }
}

Step 7: Run a Test

第七步：运行测试

Start Claude Code directly:

直接启动 Claude Code：

claude

Then say "hello" to test.

然后说 "你好" 测试。

⚠️ Important: Do not use /model minimaxai/minimax-m2.7 within Claude Code! The correct approach is to speak directly; the proxy will automatically map the request to the model you configured in the .env file.

⚠️ 重要：不要在 Claude Code 中使用 /model minimaxai/minimax-m2.7！正确方式是直接说话，代理会自动映射到你在 .env 中配置的模型。

Recommended Available Models

可用模型推荐

Based on nvidia_nim_models.json, the following models are available for free:

根据 nvidia_nim_models.json，以下模型都可以免费使用：

Model Characteristics
z-ai/glm-5.1 Strong Chinese understanding, long context
z-ai/glm4.7 Stable and reliable
minimaxai/minimax-m2.7 Fast coding speed
minimaxai/minimax-m2.5 More lightweight
moonshotai/kimi-k2-thinking Strong thinking ability
stepfun-ai/step-3.5-flash Generous free quota

Frequently Asked Questions

常见问题

Q: What is the difference between free-claude-code and CC Switch?

Q: free-claude-code 和 CC Switch 有什么区别？

Comparison Item CC Switch free-claude-code
Implementation Tool-assisted configuration Local proxy service
Model Mapping Manual model name entry Unified configuration via .env
Request Conversion Simple replacement Full Anthropic → OpenAI conversion
Tool Calling Less stable Heuristic parsing, more stable
Debugging Difficulty Harder to troubleshoot Clear terminal logs
Configuration Method GUI interface Configuration file

Why does free-claude-code have a higher success rate?

为什么 free-claude-code 成功率更高？

  1. Unified Model Name Format: CC Switch requires manual entry of model names, which is prone to errors (e.g., case sensitivity, namespaces). free-claude-code uses the .env file for unified configuration with a fixed format.

    模型名称格式统一：CC Switch 需要手动填写模型名，容易填错（如大小写、命名空间），free-claude-code 用 .env 统一配置，格式固定

  2. Complete Request Conversion: CC Switch may only perform simple model name replacement, while free-claude-code carries out full API format conversion (including thinking, tool calls, etc.).

    完整的请求转换：CC Switch 可能只做了简单的模型名替换，而 free-claude-code 做了完整的 API 格式转换（包括 thinking、tool call 等）

  3. No Residual Configuration Interference: Directly modifying settings.json is cleaner than using a tool, as CC Switch may have residual configurations that interfere.

    无残留配置干扰：直接修改 settings.json 比工具更干净，CC Switch 可能有残留配置影响

  4. Community Validation: free-claude-code is the most popular Claude Code proxy solution on GitHub, with detailed documentation and easier troubleshooting.

    社区验证：free-claude-code 是 GitHub 上最受欢迎的 Claude Code 代理方案，文档详细，问题容易排查

Q: Is it necessary to use a batch script?

Q: 一定要用批处理脚本吗？

No! After configuring settings.json, running claude in any terminal will automatically go through the proxy.

不需要！ 配置好 settings.json 后，在任意终端运行 claude 都会自动走代理。

If you don't want to manually start the proxy each time, you can create a 启动代理.bat file and double-click to run it:

如果你不想每次手动启动代理，可以创建一个 启动代理.bat 双击运行：

@echo off
cd /d D:\project2026\free-claude-code
start "NVIDIA NIM Proxy" cmd /k ".venv\Scripts\uvicorn.exe server:app --host 0.0.0.0 --port 8082"

Q: Why am I still getting the error "model may not exist"?

Q: 为什么还是报错 "model may not exist"？

Check these three points:

检查三点：

  1. Is ANTHROPIC_BASE_URL in Claude Code's settings.json correctly set to http://localhost:8082?

    Claude Code 的 settings.json 是否正确设置了 ANTHROPIC_BASE_URL 为 http://localhost:8082

  2. Is the proxy running (listening on port 8082)?

    代理是否正在运行（端口 8082 是否监听）

  3. Is the model name in the .env file correct (with namespace, e.g., z-ai/glm-5.1)?

    .env 中的模型名称是否正确（带命名空间，如 z-ai/glm-5.1）

Q: Is the free quota sufficient?

Q: 免费额度够用吗？

The NVIDIA NIM free tier allows 40 requests per minute, which is more than sufficient for daily development. For complex tasks, it is advisable to slow down the pace.

NVIDIA NIM 免费层 40 次/分钟，对于日常开发完全够用。复杂任务建议适当放慢速度。

Q: Can I use multiple models at the same time?

Q: 可以同时用多个模型吗？

Yes! In the .env file, set different models for MODEL_OPUS, MODEL_SONNET, and MODEL_HAIKU to enable a mixed usage.

可以！在 .env 中为 MODEL_OPUS、MODEL_SONNET、MODEL_HAIKU 分别设置不同的模型，实现混合使用。

Q: How do I switch models?

Q: 切换模型怎么操作？

  1. Edit D:\project2026\free-claude-code\.env

    编辑 D:\project2026\free-claude-code\.env

  2. Modify the MODEL value:

    修改 MODEL 的值：

MODEL="nvidia_nim/z-ai/glm-5.1"      # GLM-5.1
MODEL="nvidia_nim/minimaxai/minimax-m2.7"  # MiniMax M2.7
MODEL="nvidia_nim/moonshotai/kimi-k2-thinking"  # Kimi K2
  1. Restart the proxy (double-click 启动代理.bat).

    重启代理（双击 启动代理.bat）

Conclusion

总结

free-claude-code is arguably the most seamless solution for domestic users to use Claude Code for free. It perfectly preserves the powerful engineering capabilities of Claude Code while allowing us to leverage high-quality models like GLM-5.1 and MiniMax M2.7 at no cost.

free-claude-code 可能是目前国内用户免费使用 Claude Code 最丝滑的方案。它完美保留了 Claude Code 强大的工程能力，同时让我们能白嫖 GLM-5.1、MiniMax M2.7 等优质模型。

For projects of low to medium complexity, this solution is more than adequate. Why not save the subscription fee and spend it on something else?

对于中低复杂度的项目来说，这套方案已经完全够用。省下的订阅费买点啥不香呢？

常见问题（FAQ）

free-claude-code 免费使用需要什么 API？如何获取？

需要 NVIDIA NIM API Key。前往 build.nvidia.com 申请以 nvapi- 开头的 Key，免费额度每分钟 40 次请求，注册即用。

使用 free-claude-code 会失去 Claude Code 的原有功能吗？

不会。项目通过代理伪装，Claude Code 无感知，保留上下文管理、记忆系统、工具调用等全部工程能力，同时支持多模型混搭。

free-claude-code 支持哪些免费模型提供商？

支持 NVIDIA NIM（免费 40 次/分钟）、OpenRouter（含免费模型）、DeepSeek（低价稳定），以及本地 LM Studio 和 llama.ccp。

标签
DeepSeekAI大模型人工智能
← 返回文章列表
分享到：微博
下一篇
如何同时调用OpenAI、Anthropic等多个大模型？2026年Abso库使用推荐

版权与免责声明：本文仅用于信息分享与交流，不构成任何形式的法律、投资、医疗或其他专业建议，也不构成对任何结果的承诺或保证。

文中提及的商标、品牌、Logo、产品名称及相关图片/素材，其权利归各自合法权利人所有。本站内容可能基于公开资料整理，亦可能使用 AI 辅助生成或润色；我们尽力确保准确与合规，但不保证完整性、时效性与适用性，请读者自行甄别并以官方信息为准。

若本文内容或素材涉嫌侵权、隐私不当或存在错误，请相关权利人/当事人联系本站，我们将及时核实并采取删除、修正或下架等处理措施。 也请勿在评论或联系信息中提交身份证号、手机号、住址等个人敏感信息。

您可能感兴趣