一句话:GitHub上有个开源项目awesome-freellm-apis,整理了全球134个免费LLM API,390+模型,一键配置就能让Cursor、Claude Code白嫖上Claude、GPT、Gemini。我花了一上午配置完,现在我的AI编程工具完全不用充钱了。
发生了什么
免费LLM API这事,以前得自己到处找、到处注册、到处试。现在好了——一个GitHub项目把134个免费API全收在一起了。
这个项目叫awesome-freellm-apis,每天自动更新数据(截至2026年7月1日),收录了30家提供商、390+个模型。关键是它给每个API都提供了一键配置代码——你复制粘贴就能把免费模型接到Cursor、Claude Code、OpenClaw这些AI编程工具里用。
我实测了一下,5分钟配好Google Gemini 2M上下文的免费API,直接开始写代码,体验不输Claude Sonnet。关键是——完全免费。
为什么重要
1. 告别"这个模型用不起"的焦虑
Claude Opus一个月20美元,GPT-5 Pro更贵。但开源模型的质量已经追得很近了——Gemini 2.0 Flash、DeepSeek V3、Qwen3系列,在很多任务上跟付费模型差距不到10%。白嫖完全够用。
2. 国内用户终于有好用的进出口
这个项目里中国开发者能直接用的至少有5家——DeepSeek、智谱AI、硅基流动、阿里云Model Studio、ModelScope。不需要梯子,注册就能用,API兼容OpenAI格式,接入成本为零。
3. AI编程工具从此免费
Cursor试用期过了?Claude Code要充钱?别急——这些工具都支持自定义API。你把免费API填进去,它们就变成永久免费版了。这也是为什么LongCat-2.0测试版在Claude Code上月调用量排全球第二——大家都在用免费模型替代付费的。
我的判断
免费API的真正价值不是"省了20刀"——而是降低了AI切换成本。以前你锁定在Cursor的$20里不敢换,现在你可以同时用5个模型做对比:Gemini写代码快、DeepSeek逻辑好、Qwen中文强。免费让你有了选择权,这才是最大的利益。
但也要清醒:免费API有速率限制,不适合高频调用或生产环境。开发测试、个人项目、学习场景才是它的主战场。
你可以做什么
第1步:挑一个最合适的免费API
我帮你筛选了5家最适合中国开发者的,按推荐优先级排序:
Groq — 零门槛首选。无需信用卡,注册即用,30次/分钟。支持Llama 4、DeepSeek R1等开源模型,推理速度极快。注册:console.groq.com
Google Gemini — 完全免费。13个模型,2M超长上下文,15次/分钟。Gemini 2.0 Flash的代码能力能打Claude Sonnet。注册:aistudio.google.com
DeepSeek(深度求索) — 国内首选。V3.2和R1模型,128K上下文,中文能力最强。注册:platform.deepseek.com
智谱AI — 国内免费。GLM-4.7-Flash模型,200K上下文,还支持图像和视频。注册:open.bigmodel.cn
硅基流动 — 国内平台。DeepSeek R1、Qwen等开源模型,30次/分钟,注册:cloud.siliconflow.cn
第2步:配置到你的AI编程工具
以Cursor为例,打开Settings,找到Models,添加一个自定义Provider:
# Cursor添加Google Gemini免费API Base URL: https://generativelanguage.googleapis.com/v1beta/openai/ API Key: 你的Gemini API Key(从aistudio.google.com获取) Model: gemini-2.0-flash
Claude Code同理,在配置文件里加上自定义provider就行。awesome-freellm-apis项目里每个API都贴了一键配置,复制粘贴30秒搞定。
第3步:彩虹桥 —— Puter.js 零配置方案
如果你连注册API Key都嫌麻烦,还有个更狠的方案——Puter.js。它是一个"用户付费"模式的免费AI API网关,支持500+模型(GPT-5.4、Claude Sonnet 4.6、Gemini、DeepSeek全有),不需要任何API Key。
在HTML里加一行script标签就能用:
适合快速原型、前端Demo、学习实验。不适合高频调用——毕竟人家也没收你钱。
适用场景清单
✅ 适合的场景:
✅ 个人项目开发测试 —— 免费API完全够用
✅ AI编程工具白嫖 —— Cursor/Claude Code接入免费模型
✅ 学习AI开发 —— 不用充钱就能玩转多个模型
✅ 模型对比测试 —— 同时用5个模型跑同一个Prompt看效果
⚠️ 不太适合的场景:
⚠️ 生产环境高频调用 —— 速率限制会让你怀疑人生
⚠️ 需要SLA保障的企业场景 —— 免费的不保证可用性
⚠️ 处理敏感数据的场景 —— 别把机密信息发到免费API
去 GitHub 搜 awesome-freellm-apis 马上开始
参考来源:github.com/open-free-llm-api/awesome-freellm-apis(数据截至2026-07-01)| developer.puter.com(Puter.js文档,更新于2026-06-11)| console.groq.com | aistudio.google.com | platform.deepseek.com | open.bigmodel.cn | cloud.siliconflow.cn
一个更开放的问题:免费AI API越来越多,付费的还有人买吗?我的看法是——免费的是获客手段,付费的是服务保障。等你项目上了规模,你会自动付费。但在此之前,白嫖是开发者的基本权利。你觉得呢?
下一篇我准备实测几个免费聚合API平台——OpenRouter、硅基流动、Groq的横向对比,告诉你哪个延迟最低、哪个最稳定。关注「AI上手笔记」,不花冤枉钱。
你用哪个免费LLM API最多?Groq的死忠还是Gemini的粉丝?评论区说说,我挑3个点赞最高的下周单独评测。
标签: #AI工具 #免费AI #AIAPI #AI教程 #LLM #Cursor #ClaudeCode #开发者
—— AI上手笔记 · 让AI替你省时间 ——
In one line: there is an open-source project on GitHub called awesome-freellm-apis that curates 134 free LLM APIs worldwide, 390+ models, with one-click configuration that lets Cursor and Claude Code use Claude, GPT and Gemini for free. I spent a whole morning configuring it, and now my AI programming tools cost me nothing at all.
What happened
The idea of free LLM APIs used to mean hunting around, signing up everywhere and trying things one by one. That's no longer the case — a GitHub project has now collected all 134 free APIs in one place.
The project is called awesome-freellm-apis. Its data is updated automatically every day (as of July 1, 2026), covering 30 providers and 390+ models. The key thing is that it provides one-click configuration code for every API — copy, paste, and you can hook a free model into AI programming tools like Cursor, Claude Code and OpenClaw.
I tested it myself: I configured Google's free Gemini API with a 2M context window in 5 minutes and started writing code straight away, and the experience is no match for Claude Sonnet. The key point — it is completely free.
Why it matters
1. Saying goodbye to the anxiety of 'I can't afford this model'
Claude Opus costs 20 dollars a month, and GPT-5 Pro is even more expensive. But open-source models have closed the gap considerably — Gemini 2.0 Flash, DeepSeek V3 and the Qwen3 series are within 10% of paid models on many tasks. Free is more than enough.
2. Chinese users finally get good inbound and outbound options
At least 5 providers in this project are directly usable by Chinese developers — DeepSeek, Zhipu AI, SiliconFlow, Alibaba Cloud Model Studio and ModelScope. No VPN needed, sign up and use it, the APIs are OpenAI-compatible, and integration cost is zero.
3. AI programming tools become free for good
Cursor's trial period over? Claude Code asking for payment? Don't panic — all of these tools support custom APIs. Plug a free API in and they become permanently free. That is also why the LongCat-2.0 beta ranked second globally in monthly call volume on Claude Code — everyone is substituting free models for paid ones.
My take
The real value of free APIs is not 'saving 20 bucks' — it is lowering the cost of switching between AI tools. Previously you were locked into Cursor's $20 plan and afraid to move; now you can run five models side by side: Gemini for fast coding, DeepSeek for logic, Qwen for Chinese. Free gives you choice, and that is the biggest benefit. But stay clear-headed too: free APIs have rate limits and are not suitable for high-frequency calls or production. Development and testing, personal projects and learning are where they belong.
What you can do
Step 1: Pick the most suitable free API
I have filtered down 5 providers that suit Chinese developers best, in recommended order:
Groq — the zero-friction top pick. No credit card required, sign up and use it, 30 requests/minute. Supports open-source models such as Llama 4 and DeepSeek R1, with extremely fast inference. Sign up: console.groq.com
Google Gemini — completely free. 13 models, a 2M ultra-long context window, 15 requests/minute. Gemini 2.0 Flash's coding ability rivals Claude Sonnet. Sign up: aistudio.google.com
DeepSeek — the top domestic pick. V3.2 and R1 models, 128K context, and the strongest Chinese-language capability. Sign up: platform.deepseek.com
Zhipu AI — free in China. GLM-4.7-Flash model, 200K context, and it also supports image and video. Sign up: open.bigmodel.cn
SiliconFlow — a domestic platform. Open-source models such as DeepSeek R1 and Qwen, 30 requests/minute. Sign up: cloud.siliconflow.cn
Step 2: Configure it into your AI programming tool
Taking Cursor as an example, open Settings, find Models and add a custom provider:
Claude Code works the same way — just add a custom provider in the config file. The awesome-freellm-apis project includes one-click configuration for every API, so copy and paste and you are done in 30 seconds.
Step 3: The rainbow bridge — a zero-configuration option with Puter.js
If even registering for an API key feels like a hassle, there is an even better option — Puter.js. It is a free AI API gateway based on a 'user pays' model, supporting 500+ models (GPT-5.4, Claude Sonnet 4.6, Gemini and DeepSeek are all there), and it requires no API key at all.
Adding a single script tag in your HTML is enough to use it:
It suits rapid prototyping, front-end demos and learning experiments. It is not suitable for high-frequency calls — after all, they are not charging you anything.
List of applicable scenarios
✅ Scenarios that suit it:
✅ Personal project development and testing — free APIs are more than enough
✅ Free AI programming tools — hook free models into Cursor / Claude Code
✅ Learning AI development — try out multiple models without paying
✅ Model comparison testing — run the same prompt across 5 models and compare results
⚠️ Scenarios that do not suit it:
⚠️ High-frequency calls in production — the rate limits will make you question your life choices
⚠️ Enterprise scenarios requiring SLA guarantees — free services do not promise availability
⚠️ Handling sensitive data — do not send confidential information to a free API
Search for awesome-freellm-apis on GitHub and get started right away
Sources: github.com/open-free-llm-api/awesome-freellm-apis (data as of 2026-07-01) | developer.puter.com (Puter.js documentation, updated 2026-06-11) | console.groq.com | aistudio.google.com | platform.deepseek.com | open.bigmodel.cn | cloud.siliconflow.cn
A more open question: free AI APIs are growing in number — will anyone still pay for the paid ones? My view is — free is a customer-acquisition tactic, and paying is a guarantee of service. Once your project scales up, you will pay automatically. Until then, freeloading is a basic right for developers. What do you think?
In the next article I plan to put several free API aggregator platforms to the test — a head-to-head comparison of OpenRouter, SiliconFlow and Groq, to tell you which has the lowest latency and which is the most stable. Follow 'AI Hands-On Notes' and stop overpaying.
Which free LLM API do you use the most? A Groq loyalist or a Gemini fan? Tell me in the comments and I will review the 3 most-liked ones individually next week.
Tags: #AITools #FreeAI #AIAPI #AITutorial #LLM #Cursor #ClaudeCode #Developers
— AI Hands-On Notes · Let AI save you time —