
最近 Token 贵得离谱。
微软因为 API 账单太狠,把内部 Claude Code 的许可证给取消了。Uber 的 CTO 发内部备忘录说,2026 年一整年的 AI 预算,四个月就烧光了。有个独立开发者三周烧掉 1.4 亿 Token,账单近 1000 美元,发推吐槽说"Token 太贵了"。
这不是个别现象。GitHub 开始取消统一费率,全面转向按量计费。美国 AI 软件价格普涨 20% 到 37%。Anthropic 搞"20 美元底价 + 实际 Token 消耗",OpenAI 企业版全面按 Token 量挂钩。
用 AI 的成本,正在变成越来越奢侈的事情。
就在这个节骨眼上,一家叫 Agnes AI 的公司干了一件让整个圈子都炸了的事:从 2026 年 6 月 1 日起,旗下三大核心模型 API,无限期免费开放。
不是试用三天,不是免费版阉割功能,不是绑卡后偷偷扣费——是真的免费。
Agnes AI 是谁?
先说背景。Agnes AI 是新加坡的一家独立 AI Lab,全球注册用户超过 600 万。它在几个国际评测榜单上都进了前十:
• Claw-Eval(智能体评测榜):Agnes-2.0-Flash 进了 Top 10,超越了 Gemini Flash 和 MiniMax M2.7
• Artificial Analysis(图像编辑榜):Agnes-Image-2.0-Flash 进了全球前十
• Artificial Analysis(视频生成榜):Agnes-Video-V2.0 也有上榜
这不是什么野鸡公司,是正经在全球榜单上跟 OpenAI、Anthropic、Google 排在一起的。
免费的是哪三个模型?
| 模型 | 类型 | 免费前价格 | 现在 |
|---|---|---|---|
| Agnes-2.0-Flash | 文本 | 输入 $0.03/百万Token,输出 $0.15/百万Token | 免费 |
| Agnes-Image-2.0-Flash | 图像 | $3/1000张 | 免费 |
| Agnes-Video-V2.0 | 视频 | $0.3/分钟 | 免费 |
三个模态——文本、图像、视频——全部免费,无限期。同时免费开放这三个模态 API 的,Agnes 是第一家。
实际能干什么?
文本模型
Agnes-2.0-Flash 支持 1M 超长上下文,原生支持 Function Calling、Tool Use 和结构化输出。说人话就是:它不需要你额外套壳,自己就能跑 Agent,能和搜索引擎、代码执行环境无缝协同。
有人用它生成了一个完整的 React 记忆翻牌游戏,跑起来丝滑无比。还有人让它做了一个类似 X 的社交分享网站,信息流、搜索框、关注按钮、侧边导航栏全都有,点赞还能触发动画。
对于产品原型、Demo 展示,这个模型已经够用了。
图像模型
Agnes-Image-2.0-Flash 的核心能力是"按指令编辑图"。给一张图加一句话,模型就能换背景、加物体、改风格、修文字、融合多图。
有人随手丢了一张吐司的照片进去,配了一句话"为这片面包生成一张电商海报",结果渲染出来的海报可以直接挂到天猫首页。奇亚籽小麦面包、0 糖标签、暖色调光影、木质托盘,全都有了。
更关键的是,它在人像编辑时能稳定保持人脸一致。电商虚拟试穿、人物海报这些场景,不用担心把客户 P 成另一个人。
视频模型
Agnes-Video-V2.0 支持音画同步生成、首帧生视频、首尾帧生视频、多镜头切换,720P 和 1080P 随便选。
有人测了一个酒吧吉他弹唱的视频,人声、乐器、口型全部同步,跟真人拍摄区别不大。还有人测了一段街头打电话的表演片段,面部表情和情绪变化随着对话内容调整,看起来更像实景拍摄。
怎么接入你的 Agent?
Agnes AI 的 API 兼容 OpenAI 格式,接入方式很通用。官方专门写了各个工具的接入指南,我挑几个常见的:
接入 Hermes Agent
四条命令搞定:
hermes config set model.provider custom
hermes config set model.base_url https://apihub.agnes-ai.com/v1
hermes config set model.api_key YOUR_API_KEY
hermes config set model.default agnes-2.0-flash
接入 Claude Code
打开 cc-switch,右上角选 Claude CLI,点 + 号选自定义配置。API key 填进去,请求地址写 platform.agnes-ai.com,API 格式选 OpenAI Chat Completions。获取模型列表,选 agnes-2.0-flash,完
接入 OpenClaw / Claude Desktop / 其他
官方文档里都有现成教程:agnes-ai.com/doc。
注册地址:platform.agnes-ai.com,不绑卡,不排队,注册就能拿到 API key。
免费的东西靠谱吗?
我知道你在想什么——"免费的最贵"。
但 Agnes AI 的免费模式有它的逻辑。它之前就是走低价路线的,文本模型定价已经只有 DeepSeek V4 Flash 的一半,图像模型更是行业平均价的十分之一。这次全面免费,更像是把本来就很低的门槛彻底踩平。
当然,免费模式能持续多久、怎么盈利,这些都是后话。但至少在当下:
• 你不用再为 Token 账单焦虑
• 你可以在 Agent 工作流里放心跑多轮对话
• 你可以同时用文本、图像、视频三个模态,不用分三家注册
写在最后
当硅谷大厂在涨价的时候,有人在免费。
你当然可以继续给大厂交 Token 保护费,但也可以试试 Agnes AI 的免费 API。
注册地址:
platform.agnes-ai.com开发文档:
agnes-ai.com/doc
Recent Token prices have become absurd.
Microsoft canceled its internal Claude Code license due to excessive API costs. Uber's CTO issued an internal memo stating that the 2026 AI budget was exhausted within four months. A solo developer burned through 140 million tokens in three weeks, resulting in a bill of nearly $1,000, prompting them to tweet "Tokens are too expensive."
This is not an isolated phenomenon. GitHub is abandoning flat-rate pricing in favor of complete pay-as-you-go. U.S. AI software prices have increased by 20% to 37%. Anthropic introduced a "minimum $20 charge plus actual token consumption," and OpenAI Business went fully token-based.
The cost of using AI is becoming increasingly luxurious.
At this critical juncture, a company called Agnes AI did something that shook the entire community: starting June 1, 2026, their three core model APIs will be opened to the public for free, indefinitely.
This is not a three-day trial, nor a censored free version, nor a hidden subscription fee after binding a card — it is genuinely free.
Who is Agnes AI?
First, the background. Agnes AI is an independent AI lab based in Singapore with over 6 million registered users worldwide. It ranks in the top ten of several international benchmarks:
- Claw-Eval (Agent benchmark): Agnes-2.0-Flash placed in the top 10, surpassing Gemini Flash and MiniMax M2.7.
- Artificial Analysis (Image Editing): Agnes-Image-2.0-Flash is among the global top 10.
- Artificial Analysis (Video Generation): Agnes-Video-V2.0 is also listed.
This is not a fly-by-night operation — it is a respected company standing alongside OpenAI, Anthropic, and Google.
Which three models are free?
All three modalities — text, image, and video — are free, indefinitely. Agnes is the first to offer this simultaneous free access to all three APIs.
What can you actually do with these?
Text Model
Agnes-2.0-Flash supports 1M+ long context and natively handles Function Calling, Tool Use, and structured output. In layman’s terms: it doesn’t need an external wrapper; it can run agents by itself and seamlessly interact with search engines and code execution environments.
Someone successfully generated a full React memory flip game with it, running smoothly. Another created a social sharing website similar to X, complete with a feed, search bar, follow buttons, and side navigation — even the like button triggers an animation.
For product prototyping and demo presentations, this model is more than sufficient.
Image Model
Agnes-Image-2.0-Flash excels at "instruction-based image editing." Give a picture and a sentence, and the model can change backgrounds, add objects, adjust styles, edit text, and merge multiple images.
One user sent a photo of a slice of toast and asked for an e-commerce poster. The result was a high-quality poster with a wooden tray, warm lighting, and a clear product label — suitable for direct use on a marketplace.
Crucially, it maintains face consistency during portrait editing, which is essential for e-commerce virtual try-ons and portrait advertising — no more customer faces turning into random people.
Video Model
Agnes-Video-V2.0 supports synchronized audio-visual generation, single-frame-to-video, end-frame-to-video, and multi-shot switching, with options for 720P and 1080P.
A test of a bar guitarist showed perfect synchronization between guitar playing, vocals, and lip movement — indistinguishable from a real recording. Another clip of street acting showed accurate facial expressions corresponding to the dialogue, indicating high fidelity.
How to integrate into your Agents?
Agnes AI’s API is compatible with the OpenAI format, making integration straightforward. Official guides are available for various tools:
For Hermes Agent: - `hermes config set model.provider custom` - `hermes config set model.base_url https://apihub.agnes-ai.com/v1` - `hermes config set model.api_key YOUR_API_KEY` - `hermes config set model.default agnes-2.0-flash`
For Claude Code: Open cc-switch, select Claude CLI, then Custom Configuration. Enter the API key, set the request address to `platform.agnes-ai.com`, and choose "OpenAI Chat Completions" format. Retrieve models and select `agnes-2.0-flash`.
For OpenClaw / Claude Desktop / others: Refer to the official documentation at `agnes-ai.com/doc` for detailed setup instructions.
Registration is at `platform.agnes-ai.com`. No card binding, no waiting — register and get an API key immediately.
Is this free offering reliable?
You might be thinking, "Free is expensive."
However, Agnes AI’s free model has logic behind it. They have historically offered competitive pricing — text models were half the cost of DeepSeek V4 Flash, and image models were a tenth of the industry average. This full free access appears to be a move to completely remove barriers.
Of course, the sustainability of the free model and its future monetization strategy are unknown. But for now: - You no longer need to worry about token bills. - You can run multi-turn conversations in your agent workflows without hesitation. - You can use text, image, and video simultaneously without registering with three different providers.
Final Thoughts
While Silicon Valley giants are raising prices, someone is offering free access.
You can continue to pay token protection fees to the big players, or you can try Agnes AI’s free API.
Register here: `platform.agnes-ai.com`
Developer documentation: `agnes-ai.com/doc`
But Agnes AI's free model has its logic. It used to take a low-price approach, with text model pricing already half of DeepSeek V4 Flash, and image model even one-tenth of the industry average. This full-scale free offering seems more like flattening an already very low threshold.
Of course, how long the free model can sustain and how to monetize are subsequent issues. But at least currently:
• You no longer need to worry about Token bills
• You can safely run multiple rounds of dialogue in Agent workflows
• You can use text, image, and video modes simultaneously without registering with three different companies
写在最后
While Silicon Valley giants are raising prices, someone is offering it for free.
You can certainly continue paying Token fees to the giants, but you can also try Agnes AI's free API.
注册地址: platform.agnes-ai.com
开发文档: agnes-ai.com/doc