DeepSeek又调价了,我找到了8个免费薅羊毛的平台。DeepSeek raises prices again; here are 8 platforms to get freebies.


AI free token/AI免费积分

6月29日,DeepSeek调整了收费标准。

 

之前一直在降价,这次终于涨了。

 

我算了一下,一个Agent跑一个月,API费用可能几十到几百。

 

听起来不多,但对于我一个大学生来说,这也是钱啊。

 

有没有办法不花钱也能用Agent?

 

我翻了无数个帖子,踩了无数个坑,最后筛选了八个平台,注册就送免费额度,让你零成本上手Agent。

 

图片

 

先说为什么用API,不用网页端。

 

网页端免费,但没法接Agent,也不能长期记忆,只能一问一答。

 

API能接入Agent,实现自动化。

 

比如用Agent自动写日报、自动整理笔记、自动监控信息(详见我前几篇文章)。

 

而想用Agent,得先去获取API。

 

第一个平台,Agnes AI。

 

这个平台目前完全免费,无限额度。

 

多模态API,文字模型、图像模型、视频模型都能用。

 

可以接入Claude Code、Codex、Hermes。

 

我用它跑了一周Agent,一分钱没扣。

 

趁现在免费,赶紧薅。

 

图片生图模型测试

 

 

第二个平台,日日新 SenseNova。

 

商汤科技的大模型平台,有四个免费模型,最后一个是新加的。

 

SenseNova 6.7 Flash-Lite,每5小时1500次调用,支持文本对话和图像理解。

 

SenseNova U1 Fast,每5小时1500次调用,专门生成信息图,中文文字准确度很高。

 

DeepSeek V4 Flash,每5小时500次调用,支持1M上下文和工具调用。

 

Glm5.2,国产最强编程模型,但是现在官方没有透露具体额度。

 

用它跑一些简单任务,额度够用。

 

注册简单,不用绑卡。

 

图片

 

第三个平台,StepFun 阶跃星辰。

 

注册送15天免费体验,15天内用完再赠15天。

 

每邀请1位好友注册,额外获得15天,累计最高120天。

 

而且我发现它这个免费套餐还可以用自带的step-router-v1 路由转接DeepseekV4Pro,这样的话,简单任务用step模型,复杂的用DeepSeek,效率更高。

 

这个模型速度很快,单任务最高超过四百TPS,而绝大多数的推理模型都在一百TPS以下。

 

活动截至7月18日,赶紧薅。

图片

 

第四个平台,阿里云百炼。

 

新用户7000万加Token免费。

 

可用模型有通义千问、DeepSeek、GLM等多模型。

 

适合长文本处理、复杂推理、多模型切换。

 

用完即停,防止意外扣费。

 

学生注册之后,还送三百块的体验金可以用来购买各种套餐产品。

 

第五个平台,Mimo平台。

 

我现在给Hermes使用的主力模型,国产模型里排第一档。

 

注册送十块钱体验金,可以跑一个小项目,感受一下Mimo的模型。

 

可惜之前的万亿token计划结束了,我当时白嫖了几千块的套餐额度,用都用不完。

 

MIMO的TTS是免费的,可以拿来语音合成。

 

走我邀请链接可以领十元体验金和打九折:https://platform.xiaomimimo.com?ref=QZDUR3

 

第六个平台,硅基流动。

 

注册送16块钱额度。

 

适合语义搜索、知识库搭建。

 

里面有很多免费模型,我的RAG搭建用的就是里面的Embedding模型,送的免费额度根本用不完。

 

同样可以走我的邀请链接:https://cloud.siliconflow.cn/i/7cE0L6za

邀请码:7cE0L6za

 

 

最后两个平台是国产自研Agent,WorkBuddy和Qoder Work。

 

WorkBuddy注册送2000多积分,完成任务送4000多积分,每天签到150积分。

 

Qoder Work注册送2000积分,学生和老师额外送4000积分,每天签到100积分。

 

workBuddy邀请链接:https://www.codebuddy.cn/events/invite?inviteCode=z4fk31jtl

 

qoderwork邀请链接:https://qoder.com.cn/referral?referral_code=MkrgAwIyrwPOdpR05kOBWawenWcSJYN9

 

支持很多国产模型。

 

这里分享一个小技巧,闲鱼上几块钱能买workbuddy5000积分,拿来跑GLM 5.2性价比拉满。

 

适合Agent任务、自动化工作流。

 

最后总结一下。

 

日常对话用SenseNova,生图用Agnes和SenseNova,Agent用WorkBuddy。

 

最简单的是阿里云百炼,最划算的是Agnes。

 

自从接了这些平台,我现在每天用Agent自动整理笔记,一分钱没花。

 

然后我打算建个群,大家可以在里面交流AI经验,我会在里面分享我的一些ai自动化流程和各种工具教程,想进的可以加我微信,备注ai。

图片

 

On June 29, DeepSeek adjusted its pricing policy.

After continuous price reductions, the charges have finally increased.

I calculated that running an agent for a month might cost dozens to hundreds of dollars in API fees.

While this may not seem like much, for a college student like me, it’s a significant expense.

Is there a way to use agents without spending money?

I browsed numerous posts and tried many platforms. Eventually, I found eight platforms that offer free credits upon registration, allowing you to get started with agents at no cost.

图片

First, why use API instead of web interface?

The web interface is free but lacks integration with agents and long-term memory, only supporting single-question-single-answer interactions.

APIs enable connection to agents, facilitating automation.

For example, agents can be used to automatically write daily reports, organize notes, and monitor information (detailed in my previous articles).

To use an agent, you first need to obtain an API.

The first platform is Agnes AI.

This platform is currently completely free with unlimited quotas.

It supports multimodal APIs, including text models, image models, and video models.

You can connect to Claude Code, Codex, and Hermes.

I ran an agent on it for a week without being charged a dime.

Take advantage of the free period now.

Image model test图片

The second platform is SenseNova by Megvii.

This large model platform offers four free models, with the most recent addition being:

SenseNova 6.7 Flash-Lite, which allows 1500 calls every 5 hours for text conversation and image understanding.

SenseNova U1 Fast, which allows 1500 calls every 5 hours, specializing in generating highly accurate Chinese infographics.

DeepSeek V4 Flash, which allows 500 calls every 5 hours, supporting 1M context length and tool calling.

GLM 5.2, the strongest Chinese programming model, though its specific quota is currently undisclosed by the official.

Its quota is sufficient for running simple tasks.

Registration is straightforward and does not require a credit card.

图片

The third platform is StepFun.

This platform offers a 15-day free trial, with the option to extend for another 15 days upon completion.

Referring one friend adds 15 days, allowing for a maximum of 120 days.

I also discovered that their free plan supports routing to DeepSeek V4 Pro via the built-in step-router-v1, making it cost-effective for simple tasks while still having the option to switch to the more expensive DeepSeek for complex ones.

The model’s speed is remarkable, reaching a maximum of 400 TPS per task, while most reasoning models hover around 100 TPS.

The promotion ends on July 18, so act quickly.

图片

The fourth platform is Alibaba Cloud Pait.

New users receive 7000 million token free credits.

Available models include Tongyi Qianwen, DeepSeek, GLM, and others.

It is suitable for long-text processing, complex reasoning, and model switching.

Services stop immediately after use, preventing unexpected charges.

Students who register also receive a 300 yuan experience coupon to purchase various service packages.

The fifth platform is the Mimo platform.

This is my primary model for running Hermes, ranking among the top for Chinese models.

Registration grants 10 yuan in experience coupons, sufficient to run a small project and test the model.

Unfortunately, their previous “trillion token” plan has ended; I had previously taken advantage of it to get thousands of yuan in free credits.

Mimo’s TTS service is free and can be used for speech synthesis.

Use my referral link to get 10 yuan in credits and a 10% discount: https://platform.xiaomimimo.com?ref=QZDUR3

The sixth platform is NVIDIA Ising-Calibration-1.5.

Registration grants 16 yuan in credits.

It is suitable for semantic search and knowledge base construction.

There are many free models available; I used the embedding model for my RAG setup, and the provided free credits were more than sufficient.

Similarly, you can use my referral link: https://cloud.siliconflow.cn/i/7cE0L6za

Invitation code: 7cE0L6za

The final two platforms are homegrown Chinese agents: WorkBuddy and Qoder Work.

WorkBuddy grants over 2000 points upon registration and awards an additional 4000 points for completing tasks, with a daily login bonus of 150 points.

Qoder Work grants 2000 points upon registration, with an extra 4000 points for students and teachers, and a daily login bonus of 100 points.

WorkBuddy referral link: https://www.codebuddy.cn/events/invite?inviteCode=z4fk31jtl

Qoder Work referral link: https://qoder.com.cn/referral?referral_code=MkrgAwIyrwPOdpR05kOBWawenWcSJYN9

They support various Chinese models.

Pro-tip: You can buy 5000 WorkBuddy points for a few dollars on Xianyu, making it highly cost-effective for running GLM 5.2.

They are suitable for agent tasks and workflow automation.

To summarize:

Use SenseNova for daily conversations, Agnes and SenseNova for image generation, and WorkBuddy for agents.

Alibaba Cloud Pait is the simplest, while Agnes offers the best value.

Since connecting with these platforms, I have been automating my note-taking daily without spending a penny.

I plan to create a group for exchanging AI experiences. I will share my automation workflows and tool tutorials there. If you’re interested, add me on WeChat with a note indicating “AI.”

图片

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