这款免费模型治好了我的token焦虑This Free Model Cured My Token Anxiety


AI free token/AI免费积分

AI 工具 · 白嫖指南

文本、图片、视频全免费,还能接到 Claude CLI 上用——

这波操作不学一下,真的亏。

2026 / 06

ONE

那天,几个大模型坐在一起聊天

某个聚会上,一群大模型凑到了一块儿。场面一度非常装逼。

Claude 先开口了,翘着二郎腿:"我推理能力强,代码写得一流,连 Opus 都敢叫自己 Opus,你们懂这个含金量吗?"

GPT 在旁边笑了笑:"行了行了,生态我最全,插件我最丰富,用户基数你们加起来都不够我看的。"

Gemini 推了推眼镜:"我背靠谷歌,多模态那都是基操,搜索更不用说了。

千问、kimi、glm、deepseek等一众互相看了一眼,说:"我俩国内市场熟……"

大家就这么聊着,互相吹着,气氛热烈。

角落里蹲着一个瘦小小身影,一直没说话。

等大家吹得差不多了,角落里那个小家伙怯生生地举了举手:

"我免费。"

场面一下子安静了。

Claude 干咳一声:"我是说,我的上下文窗口……"

"我免费。"

GPT 不服气:"不是,你到底听没听明白,我在说我的插件生——"

"我免费。"

Gemini 试图找回主场:"我们有 TPU,有搜索,有——"

"我免费。"

全体沉默。

那个小家伙看着大家,补了一刀:"文本、图片、视频,都不要钱。"

"散会。"Claude 站起来走了。

这个小家伙,就是 Agnes。

好了段子讲完,说正经的。

其实很多朋友可能还没关注到 Agnes 这个模型。简单介绍一下:Agnes 是 Sapiens AI 推出来的一系列模型,目前文本、图片、视频三条线全都免费。

先说文本模型 Agnes 2.0 Flash。这个东西不弱,在 Claw-Eval 基准测试里 Pass³ 得分 60.9%,通用排行榜排第 9,跟主流大模型放在一起完全不虚。而且它的上下文窗口有 512K,单次最大输出 65.5K,工具调用、多轮对话、代码生成、图像理解这些全支持,接口兼容 OpenAI 格式。说白了,你平时拿 Claude、GPT 干什么,拿它基本也能干,还不花钱。

图片

 

然后是图片模型 Agnes Image 2.1 Flash,支持文生图和图生图,针对那种信息密度高、细节多的画面做过专门优化。想做个海报草图、产品展示图、社交媒体封面之类的,丢个提示词过去就行。图生图还能尽量保持原图的构图和主体结构,改风格不改布局,挺实用的。

图片

 

这是用一句简单提示词“一个小孩儿坐在沙滩”生成的图片,大家感受一下。

图片

 

最后是视频模型 Agnes Video V2.0,支持文生视频、图生视频、多图生成和关键帧动画。最长能到 18 秒左右,横屏竖屏都能调,帧率和分辨率也可以自己设。适合拿来生成一些短视频素材、产品演示、角色动画之类的内容。

图片

我使用这个模型的图生视频把上面的图生成的视频效果如下:

这款免费模型治好了我的token焦虑
已关注
关注
重播 分享 赞
 
AI数字生命体
 
 
 

 

 

0/0

00:00/00:03
 
 
进度条,百分之0
 
 
00:00
/
00:03
00:03
倍速
全屏

 

倍速播放中
 
 

继续观看

这款免费模型治好了我的token焦虑

 
转载
,
这款免费模型治好了我的token焦虑
AI数字生命体
已同步到看一看写下你的评论
 

 

 
 
视频详情

 

三个模型的价格,我直接说重点:

• 文本模型:输入输出目前都是 0 元 / 百万 token

• 图片模型:0 元 一张

• 视频模型:0 元 一秒

就是白给。这你受得了吗。

适合什么人用呢?个人开发者想做点小工具又不想每个月付 API 账单的、学生党做项目没有预算的、小团队做原型验证不想一上来就烧钱的、自媒体想做点配图配视频素材的——差不多就是这些场景。当然,如果是那种对稳定性、延迟有极端要求的生产级大项目,你可能需要自己评估一下。

总之,免费的,试试又不亏。都说免费的是最贵的,也期待他们后续的商业模式是怎样的,但是现在是免费的是最重要的。

说到这儿,如果你平时在用 Claude CLI 写代码,那接下来这部分就是给你的——怎么把 Claude CLI 接到 Agnes 上,享受免费算力。

TWO

用 CC-Switch 把 Claude CLI 接到 Agnes 上

刚好官方给了文档,我照着走了一遍,确实能跑通。趁还记得,整理一份给大家。不得不赞一下,官方文档写的很详细。

官方文档地址:https://agnes-ai.com/zh-Hans/docs/overview

你需要提前准备好的东西:

• Claude CLI(已安装)

• CC-Switch(已安装)

• 一个 Agnes AI 的 API Key

• 网络能正常访问 Agnes 的 API 网关

1. 先搞到 API Key

去 Agnes AI 官网 platform.agnes-ai.com,注册登录之后进 API Key 页面创建一个 Key,复制下来。

2. 打开 CC-Switch,选对模式

启动 CC-Switch,顶部工具栏选 Claude。这个选项就是专门给 Claude CLI 做模型代理用的。

图片

3. 新增一个提供商

点右上角加号 → 提供商类型选 Claude → 选 Custom Provider(自定义提供商)。

图片

 

4. 填 API Key

把刚才复制的 Key 粘贴进去。一般不用加 "Bearer" 前缀,除非工具明确说了要完整 Authorization Header。

5. 配请求地址

请求 URL 填:

https://apihub.agnes-ai.com/v1

CC-Switch 会把 Claude CLI 的请求通过这个地址转发到 Agnes 的网关。

图片

 

6. 选 API 格式

打开高级选项,API 格式选 OpenAI Chat Completions,Agnes 的接口兼容 OpenAI 格式。

7. 认证字段用默认的

不用动。使用默认 ANTHROPIC_AUTH_TOKEN。

图片

8. 拉模型列表,做映射

点 Fetch Model List 看看能不能连上。拉成功之后,把 Claude CLI 的模型映射到 Agnes:

• Sonnet → agnes-2.0-flash

• Opus → agnes-2.0-flash

• Haiku → agnes-2.0-flash

想用其他模型也可以按需选。

图片

9. 加兼容性配置(别漏了)

在配置json里加上这段:

{

"allowed_openai_params":["thinking", "context_management"],

  "litellm_settings": {

    "drop_params": true

  }

}

这步很重要,作用是自动过滤掉不兼容的参数,避免请求报错。

图片

 

10. 保存配置

确认没填错之后点 Save。保存成功的话,提供商列表里会出现 Agnes。

11. 把路由打开

左上角设置 → 路由 → 开启本地路由 → 把 Claude 路由开关打开。

图片

12. 启用 Agnes

回到提供商列表,找到 Agnes,点 Enable。到这步就通了,Claude CLI 的请求会走 CC-Switch 转发到 Agnes 上。

图片

 

13. 验证一下

打开 Claude CLI 随便聊两句,返回正常就是配好了。看看Agnes账单页是不是已经有消费了,如果产生账单了证明切换成功了,可能有一点点的延迟。

图片

 

THREE

万一不顺利怎么办

按这几个方向排查:

• 拉不到模型列表? 先看 API Base URL 有没有写对(apihub.agnes-ai.com/v1),再确认 API Key 还有没有效。

• 认证失败? 多数是 Key 填错了,检查有没有多复制空格。再说一次,一般不用自己加 "Bearer"。

• 请求报错或参数不兼容? 回第九步,那段 JSON 加了吗,drop_params 开了吗。

• Claude CLI 没走 Agnes? 本地路由和 Claude 路由,两个开关都得开着。

• 模型返回不太对? 检查一下模型映射,推荐先用 agnes-2.0-flash 试试。

整体流程不算复杂,照着走一遍十分钟不到应该能搞定。免费的文本、图片、视频加上 Claude CLI 白嫖方案,这波怎么说都不亏。

FOUR

其他 agent 也都支持

我用的别的 agent 怎么办,没关系,都支持。官方文档找到自己在用的agent按照官方文档配置一下。

图片

 

 

觉得有用?

分享给也在折腾 AI 工具的朋友

有问题欢迎留言交流

 

AI Tools · Free-Loading Guide

Text, images and video all free — and you can even plug it into Claude CLI.

You'd miss out not learning this trick.

2026 / 06

ONE

That day, several large models sat down for a chat

At some gathering, a group of large models came together. The atmosphere was extremely pretentious for a while.

Claude spoke first, crossing its legs: 'My reasoning is strong, my coding is first-rate, I dare call myself Opus even — do you guys understand what that's worth?'

GPT laughed beside it: 'Alright, alright, I have the fullest ecosystem and the richest plugins, none of you add up to my user base.'

Gemini adjusted its glasses: 'I have Google behind me, multimodal is basic for me, and search needs no explanation.

Qwen, kimi, glm and deepseek exchanged a glance and said: 'We two know the domestic market…'

So they chatted along, boasting at each other, the atmosphere warm.

In the corner squatted a small thin figure, saying nothing the whole time.

Once everyone had finished boasting, that little fellow in the corner timidly raised a hand:

'I'm free.'

The room suddenly went quiet.

Claude cleared his throat: 'I mean, my context window…'

'I'm free.'

GPT was unconvinced: 'No, did you even understand me, I was talking about my plug-in ecosys—'

'I'm free.'

Gemini tried to win the room back: 'We have TPUs, we have search, we have—'

'I'm free.'

Everyone fell silent.

That little fellow looked at everyone and added the finishing blow: 'Text, images, video — all free.'

'Meeting adjourned.' Claude stood up and left.

That little fellow was Agnes.

All right, the joke's over. Let's get serious.

Actually many friends may not have noticed the Agnes model yet. A quick intro: Agnes is a family of models from Sapiens AI, and right now all three lines — text, image and video — are free.

First the text model Agnes 2.0 Flash. This thing is no pushover: it scored 60.9% Pass³ on the Claw-Eval benchmark and ranks 9th on general leaderboards, holding its own against mainstream large models. Its context window is 512K, single-call max output 65.5K, with tool calling, multi-turn conversation and code

图片

Then the image model Agnes Image 2.1 Flash, supporting text-to-image and image-to-image, specially optimised for frames with high information density and rich detail. Want a poster sketch, a product display image or a social media cover? Just throw a prompt at it. Image-to-image also preserves the original composition and subject structure as much as possible — change the style without changing the layout, which is quite practical.

图片

This image was generated with the single simple prompt 'a child sitting on the beach' — see for yourself.

图片

Finally the video model Agnes Video V2.0, supporting text-to-video, image-to-video, multi-image generation and keyframe animation. It goes up to about 18 seconds, handles both landscape and portrait, and frame rate and resolution are configurable. Good for generating short video material, product demos, character animation and the like.

图片

Here's the video I generated from the image above using this model's image-to-video:

Following

Follow

Replay Share Like

Close

Watch more

More

Exit fullscreen

AI Digital Life Form has followed

Share video

, duration 00:03

0/0

00:00/00:03

Switch to landscape mode

Continue playing

Progress bar, 0 percent

Play

00:00

00:03

00:03

Playback speed

Fullscreen

Speed playing

0.5x 0.75x 1.0x 1.5x 2.0x

Ultra HD Smooth

Continue watching

This free model cured my token anxiety

Watch more

Repost

,

This free model cured my token anxiety

AI Digital Life Form has followed

Share Like Recommend

Synced to Moments — write your comment

Video details

The prices of the three models, let me jump to the point:

• Text model: 0 yuan per million tokens for both input and output

• Image model: 0 yuan per image

• Video model: 0 yuan per second

Literally free. Can you handle that?

Who is it for? Individual developers who want to build small tools without paying an API bill every month, students doing projects with no budget, small teams prototyping without burning money upfront, and self-media creators who need images and video material — basically these scenarios. Of course, if it is a production-scale project with extreme demands on stability and latency, you may want to assess it yourself.

Anyway, it's free — trying costs you nothing. People say the free stuff ends up most expensive, and I look forward to seeing what their business model becomes, but for now being free is what matters.

Speaking of which, if you use Claude CLI for coding, the next part is for you — how to connect Claude CLI to Agnes and enjoy free compute.

TWO

Connecting Claude CLI to Agnes with CC-Switch

Happily the official docs exist, and I followed them through — it genuinely works. While I still remember, let me put it together for everyone. Credit where due, the official docs are very detailed.

Official docs address: https://agnes-ai.com/zh-Hans/docs/overview

What you need to prepare in advance:

• Claude CLI (installed)

• CC-Switch (installed)

• An Agnes AI API Key

• Network access to the Agnes API gateway

1. Get an API Key first

Go to the Agnes AI site platform.agnes-ai.com, register and log in, then create a Key on the API Key page and copy it.

2. Open CC-Switch and pick the right mode

Launch CC-Switch and select Claude in the top toolbar. This option exists specifically for proxying models for Claude CLI.

图片

3. Add a provider

Click the plus in the top right → provider type Claude → choose Custom Provider.

图片

4. Fill in the API Key

Paste the Key you copied. Generally you do not need to add a "Bearer" prefix unless the tool explicitly requires a full Authorization Header.

5. Configure the request address

Fill in the request URL:

https://apihub.agnes-ai.com/v1

CC-Switch forwards Claude CLI requests through this address to the Agnes gateway.

图片

6. Choose the API format

Open advanced options and set the API format to OpenAI Chat Completions — the Agnes interface is OpenAI-format compatible.

7. Leave the auth field at default

Do not change it. Use the default ANTHROPIC_AUTH_TOKEN.

图片

8. Fetch the model list and map it

Click Fetch Model List to see if the connection works. Once fetched successfully, map Claude CLI's models to Agnes:

• Sonnet → agnes-2.0-flash

• Opus → agnes-2.0-flash

• Haiku → agnes-2.0-flash

You can pick other models as needed.

图片

9. Add the compatibility config (do not skip this)

Add this block to the config JSON:

{

"allowed_openai_params":["thinking", "context_management"],

"litellm_settings": {

"drop_params": true

}

}

This step matters — it automatically filters out incompatible parameters and avoids request errors.

图片

10. Save the config

After confirming nothing is wrong, click Save. If the save succeeds, Agnes appears in the provider list.

11. Turn on routing

Top-left Settings → Routing → enable local routing → turn on the Claude routing switch.

图片

12. Enable Agnes

Return to the provider list, find Agnes and click Enable. At this point you are connected — Claude CLI requests will go through CC-Switch to Agnes.

图片

13. Verify it

Open Claude CLI and chat a couple of lines — normal responses mean it is configured. Check whether the Agnes billing page shows usage; if a charge appears the switch succeeded, there may be a slight delay.

图片

THREE

What if it doesn't go smoothly

Troubleshoot along these lines:

• Can't fetch the model list? First check the API Base URL is right (apihub.agnes-ai.com/v1), then confirm the API Key is still valid.

• Authentication failure? Usually a mistyped Key — check for extra spaces. Again, generally you do not add "Bearer" yourself.

• Request errors or incompatible parameters? Go back to step 9 — did you add that JSON, is drop_params enabled?

• Claude CLI not routing through Agnes? Both switches — local routing and Claude routing — must be on.

• Model output looks off? Check the model mapping — try agnes-2.0-flash first.

The overall flow is not complicated; following it should take under ten minutes. Free text, images and video plus the Claude CLI freeloading scheme — however you slice it, this is a win.

FOUR

Other agents are supported too

What if you use a different agent? No problem, all are supported. Find your agent in the official docs and configure it accordingly.

图片

Found this useful?

Share it with friends who are also tinkering with AI tools

Questions welcome in the comments

This Free Model Cured My Token Anxiety

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