分享目前能免费使用AI的agent软件和免费的api key,赶紧薅羊毛!Current free AI agent software and API keys — grab them while they’re available!


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/yg1S_YJbTxBnX7G_NDFnkw

如果您不愿意本站收录,请来信告知:13951431913@139.com 或在右方提交评论。

本文分为两类,一类是免费agent软件,新手老手都适用,不需要繁琐的操作,下载安装软件+注册账号就能用。
另一类是免费的api key,可以接入你的claude code、opencode等agent软件里面用。
都是免费的,确实一分钱不用花。

先说新手老手都适合的免费agent软件,下载安装+注册账号就能用。

阿里的Qoder Work现在送一个月会员,里面有Qwen3.7-Max和Qwen3.7-Plus这两个旗舰模型,晚上10点到第二天早上8点是打折时段,至少这一个月内够你用。

可以用我的邀请码注册

https://qoder.com.cn/referral?referral_code=m74k4b1vywCOQptI8lPNFRKqH2mO49gP 你的好友邀请你体验 QoderWork 中国版,阿里巴巴出品的桌面端智能工作助手,立即注册并下载客户端领 2,000 积分,完成学生/教师认证再领 4,000 积分。

图片
图片

豆包电脑版的办公任务现在也有agent能力了,也可以拿来用,但免费额度有点少。
图片
图片

字节的TRAE Work也免费。它原本是给程序员写代码用的,现在出了个"Work模式",普通人也能用大白话让它干活。热门模型偶尔要排队,就算排队也不需要等太久,但用一会会告诉你免费额度没了。如果你不指定某个模型,而是用auto自动选择模型,几乎不排队,免费额度还挺多的。

图片
图片
腾讯workbuddy

7月6日腾讯发布新模型混元HY3,顺便在它的AI agent软件WorkBuddy里免费使用,你只需要下载这个软件就能用了免费的HY3大模型了,这个大模型我看了下测评,还是不错的。

图片
图片
opencode是非常优秀的agent软件,开源+免费,里面有3个大模型目前是免费的,分别是DeepSeek V4 flash、mimo V2.5、HY3,其中mimo V2.5是多模态,所以你可以让V2.5负责多模态,其他两个模型当主力干活。
唯一要注意的是,DeepSeek V4 flash和mimo V2.5的上下文只有200K,不是满血的1M,HY3的上下文是256K。
这个免费也是有限制的,但如果你的使用强度不大,还是够用的。
图片
图片
minimax也有自己的code软件,去minimax官网下载就行了,有一些免费额度,可以用minimax M3,我只是试用了一下,不知道具体的免费额度能有多少。
图片
图片

 

Meituan CatPaw,美团也有自己的agent软件,除了美团自己的大模型外,还有GLM5.2这样的广受欢迎的大模型。

图片
图片
图片

 

tabbit浏览器,一个AI浏览器,里面也有各种大模型,包括GLM5.2、qwen3.7max、DeepSeek V4 pro。

设置为默认浏览器就可以解锁pro会员。

图片

 

图片

 

说完了免费的agent软件,接下来说说免费的AI api key,这个可能是更多人需要的。

opencode zen

https://opencode.ai/zh/zen

注册就能用,DeepSeek V4 flash和mimo V2.5免费,但上下文只有200K,都免费了,你就别要求那么高了。另外还有HY3目前也是免费的,256K上下文
商汤日日新

https://www.sensenova.cn/token-plan

可以免费用DeepSeek V4 flash和GLM5.2
但我觉得GLM5.2很难用,其他用过的朋友也这么说,所以你就当这是一个DeepSeek V4 flash的免费大模型吧,而且是100万上下文的。
图片
美团的大模型longcat-2.0

https://longcat.chat/platform/usage

注册认证后就送1000万token,而且缓存命中不计入,意味着只要缓存命中率90%以上,就能用一亿多token。很多人实测下来都能用一亿多token,还是挺便宜的。
图片
如果觉得好用,还有个9.9元5000万token的活动,同样的缓存命中免费,意味着可能会超过五亿token。
图片
阶跃星辰

https://platform.stepfun.com/?invite_code=BJFAKVRC

推荐用
step-3.7-flash
step-router-v1
多模态的快速大模型,拿来agent还是不错的
图片
英伟达

https://build.nvidia.com/models

英伟达的免费大模型,包括GLM5.2、minimax M3、kimi2.6、DeepSeek。但可能有量化,仅测试用还不错,不建议作为主力,而且热门模型会很卡。


 

agnes-ai,新加坡的一家公司,多模态大模型

图片

 

小米mimo,邀请码注册有10块钱,用mimoV2.5这个多模态大模型的话估计能用个2亿token。但考虑到上面有个opencode zen的免费V2.5大模型,这个就一般了,但都是免费的,能用先用着。

 

Source: https://mp.weixin.qq.com/s/yg1S_YJbTxBnX7G_NDFnkw

If you do not wish this site to be included, please contact us at 13951431913@139.com or submit a comment on the right.

This article is divided into two categories: free agent software, suitable for both novices and experts, requiring no complex operations, just download, install, and register to use.

The other category is free API keys that can be integrated into agent software like Claude Code or OpenCode.

All are free, absolutely no money required.

First, a free agent software suitable for both beginners and experts — just download, install, and register to get started.

Ali's Qoder Work is currently offering a free month of membership, which includes Qwen3.7-Max and Qwen3.7-Plus flagship models. There is a discount period from 10 PM to 8 AM, providing sufficient use for the month.

You can use my referral code to register: https://qoder.com.cn/referral?referral_code=m74k4b1vywCOQptI8lPNFRKqH2mO49gP. Your friend invites you to experience QoderWork China, Alibaba's desktop-based smart work assistant. Register and download the client to receive 2,000 points, and complete student/teacher authentication for an additional 4,000 points.

图片

图片

图片

图片

图片

图片

图片

图片

图片

图片

图片

图片

图片

图片

Minimax also has its own code software. Download it from the Minimax website to access free quotas for the Minimax M3 model. Specific free limits are unknown.

图片

图片

图片

图片

图片

图片

Tabbit Browser: An AI browser that includes various models like GLM5.2, Qwen3.7Max, and DeepSeek V4 pro. Setting it as the default browser unlocks pro membership.

图片

图片

图片

OpenCode Zen: https://opencode.ai/zh/zen. Register to use free models like DeepSeek V4 flash (200K context) and mimo V2.5. HY3 is also free at 256K context.

SenseNova: https://www.sensenova.cn/token-plan. Offers free usage of DeepSeek V4 flash and GLM5.2. However, GLM5.2 is reported to be difficult to use, so it is best treated as a source for free DeepSeek V4 flash with a 1M context window.

图片

图片

图片

图片

Stepfun: https://platform.stepfun.com/?invite_code=BJFAKVRC. Recommended models: step-3.7-flash and step-router-v1, which are multi-modal and suitable for agent use.

图片

NVIDIA: https://build.nvidia.com/models. Offers free models including GLM5.2, Minimax M3, Kimi2.6, and DeepSeek. These may be quantized and are best used for testing rather than primary work, as popular models can be slow.

Agnes-AI: A Singapore-based company offering multi-modal models.

图片

Xiaomi Mimo: Register with an invitation code for a 10 RMB credit, which may support up to 200 million tokens for the MimoV2.5 model. However, since OpenCode Zen already offers a free V2.5, this option is less advantageous.

图片

美团的大模型longcat-2.0

https://longcat.chat/platform/usage

After registration and verification, you’ll receive 10 million tokens — and crucially, cache hits are not counted against your quota. So if your cache hit rate exceeds 90%, you can effectively use over 100 million tokens. Many users have confirmed this, making it quite cost-effective.

图片

If you like the service, there’s a promotional offer: for 9.9 yuan, you get an additional 50 million tokens, again with free cache hits — meaning you could end up with over 500 million tokens.

图片

阶跃星辰

https://platform.stepfun.com/?invite_code=BJFAKVRC

推荐用

step-3.7-flash

step-router-v1

A fast, multimodal large model — perfect for agent applications.

图片

英伟达

https://build.nvidia.com/models

NVIDIA’s free model suite includes GLM5.2, minimax M3, kimi2.6, and DeepSeek. However, these may be quantized (compressed), and while good for testing, they aren’t recommended as primary models — especially since popular ones can get bottlenecked.

agnes-ai — a Singapore-based company offering a multimodal large model.

图片

Xiaomi’s MiMO — register with an invite code and get 10 yuan credit. Using the MiMO V2.5 multimodal model might yield ~200 million tokens, but since OpenCode ZEN offers a free V2.5 model, MiMO is more of a secondary option — though still free and usable.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry