Hermes 接入 OpenRouter,3 分钟白嫖 23 个免费大模型Hermes Connects to OpenRouter, 3 Minutes to Freeload 23 Free Large Models


AI free token/AI免费积分

用 Hermes Web UI 一键接入 OpenRouter,23 个免费模型随你挑。适合想低成本玩多模型的开发者和 AI 爱好者

想用大模型,但一看 API 价格就劝退?

GPT-5 系列按 token 计费,Claude Opus 跑一个长对话几块钱就没了。很多人想同时试几个模型做对比,光想想账单就打消了念头。

其实有个办法:OpenRouter 上有 23 个模型完全免费,不收一分钱。DeepSeek V4 Flash、Gemma 4、Nemotron 3 Super 这些能打的模型都在里面。一个 API Key 全部搞定,3 分钟就能配好。

图片

OpenRouter 是什么?免费模型到底有哪些

OpenRouter 是一个 AI 模型聚合平台。你可以理解成模型界的「应用商店」。一个 API Key,能调用几百个不同厂商的模型。不用每个厂商单独注册。

它最香的地方在于:有一批模型完全免费,不按 token 计费,不收月租。根据 OpenRouter 官方 Free Models 页面(2026 年 5 月数据),当前免费模型包括:

先看通用对话类。Owl Alpha — 105 万 token 上下文,原生支持工具调用,Agent 工作负载首选

  • • OpenAI gpt-oss-120b — OpenAI 开源的 117B MoE 模型,单卡 H100 就能跑
  • • Z.ai GLM 4.5 Air — 智谱的轻量旗舰,支持「思考模式」切换
  • • DeepSeek V4 Flash — 284B MoE,100 万 token 上下文,速度型选手

编程方向也有不少。Poolside Laguna M.1 — 专为代码 Agent 优化,128K 上下文

  • • Baidu CoBuddy — 百度千帆的代码模型,支持 65K 输出
  • • NVIDIA Nemotron 3 Super — 120B 混合 MoE,编程榜单排名前五

多模态方向,Google Gemma 4 31B 支持文本+图像输入(256K 上下文),Nemotron 3 Nano Omni 更是覆盖文/图/视频/音频全模态。

图片

还有 gpt-oss-20b、Nemotron Nano 9B、Gemma 4 26B 等轻量模型。加起来十几个厂商、覆盖对话/代码/多模态三大场景。

免费的代价是什么?部分模型会记录你的 prompt 用于改进,请求高峰时可能排队。但日常使用完全够用,后面实测会详细说。

3 分钟注册 + 拿到 API Key

注册流程极其简单,不需要信用卡。

第一步:注册 OpenRouter 账号

打开 openrouter.ai,点右上角 Sign Up。支持 Google 一键登录,也可以用邮箱注册。填完邮箱、验证,几十秒搞定。

图片

第二步:创建 API Key

登录后,点右上角头像,选 Profile → API Keys。页面中间有个「Create Key」按钮,点一下,给 Key 起个名字(随便写,比如「Hermes」),然后复制保存。

这个 Key 只显示一次,丢了就得重新创建。建议存到密码管理器或者记事本里。

到这里,OpenRouter 的部分就结束了。接下来把它接到 Hermes 里。

Hermes Web UI 一键导入,23 个模型一次到位

Hermes 是一个开源的 AI Agent 框架,它有个 Web UI 界面,可以管理多个模型渠道。用它来对接 OpenRouter,配一次就能用所有免费模型。

第一步:打开 Hermes Web UI,进入渠道管理

在左侧菜单找到「渠道」(Channels),点「添加渠道」。

图片

第二步:选择 OpenRouter,填入 API Key

渠道类型选 OpenRouter。把刚才复制的 API Key 粘贴进去,Base URL 保持默认就行(通常是 openrouter.ai/api/v1)。点确认保存。

第三步:验证连接

保存后,Hermes 会自动拉取 OpenRouter 的可用模型列表。连接成功后,渠道状态会变成绿色。模型数量显示为 23 或更多,取决于 OpenRouter 当前的免费模型数量。

到这里就配完了。从打开网页到模型可用,3 分钟足够。

接下来可以直接在 Hermes 的对话界面里选模型聊天。所有免费模型都在下拉菜单里,想用哪个点哪个。

实测:聊天、代码、PPT,免费模型到底能干啥

光说能用不行,得看看实际效果。我在 Hermes 里挑了几个场景做了测试。

场景一:日常对话

随便聊了几轮,问了几个常识性问题和一些需要推理的问题。DeepSeek V4 Flash 和 GLM 4.5 Air 的回复速度最快,基本秒回。Owl Alpha 在复杂推理上表现不错,但偶尔会慢一两秒。

日常聊天、问问题、写邮件这些场景,免费模型完全够用。

图片

场景二:代码生成

让它写了一个动态网页(带 CSS 动画的那种)。Poolside Laguna M.1 的代码质量出乎意料,一次生成的代码基本能跑,细节处理也到位。gpt-oss-120b 写代码也很稳,毕竟是 OpenAI 的开源模型。

写简单脚本、前端页面、数据处理这些日常编码任务,免费模型胜任没问题。复杂的多文件项目可能还是得上付费模型。

场景三:PPT / Skill 调用

Hermes 支持 skill(技能文档)功能,可以给模型加上特定能力。测试了 PPT 生成的 skill,让模型根据主题自动生成幻灯片内容。效果中规中矩,结构清晰,但排版细节需要手动调整。

日常使用完全够用,代码场景表现亮眼。复杂任务还是得上付费模型。 免费嘛,还要啥自行车。

几个要注意的坑

用了一段时间,踩过几个坑,提前说一下。

免费模型有请求频率限制。 高峰期(国内晚上 8-11 点)部分热门模型可能需要排队,响应变慢。如果遇到超时,换个模型试试,通常能解决。

部分模型需要加 :free 后缀。 在某些客户端里调用 OpenRouter 的免费模型,模型名后面要加 :free 才能命中免费额度。Hermes Web UI 已经帮你处理好了这个细节,不用手动加。

免费模型会记录 prompt。 OpenRouter 官方说明:部分免费模型的提供商可能会记录你的输入用于模型改进。敏感数据不要走免费模型,用付费模型或者本地部署更安全。

模型列表会变动。 OpenRouter 的免费模型不是固定的,会根据合作方调整增减。建议每隔一两个月去 openrouter.ai/collections/free-models 看一眼最新列表。

图片

一个 API Key、3 分钟配置,就能白嫖十几个厂商的免费大模型。对于个人开发者和 AI 爱好者来说,这是目前性价比最高的多模型体验方案。

下次想试新模型又心疼钱的时候,花 3 分钟走一遍上面的流程。OpenRouter 的免费模型列表会变,但「一个 Key 用所有模型」的思路不会过时。


参考来源

  • • OpenRouter Free Models 官方页面:openrouter.ai/collections/free-models
  • • OpenRouter 官方文档:openrouter.ai/docs
  • • Hermes Web UI GitHub 仓库:github.com/EKKOLearnAI/hermes-web-ui

 

Connect OpenRouter to the Hermes Web UI with one click and choose from 23 free models. Perfect for developers and AI enthusiasts who want to try many models at low cost.

You want to use a large model, but the API pricing puts you off?

GPT-5 series is billed by token, and a long conversation on Claude Opus burns a few dollars. Many people want to try several models side by side for comparison, and the mere thought of the bill kills the idea.

There is actually a solution: OpenRouter has 23 completely free models, costing nothing at all. DeepSeek V4 Flash, Gemma 4, Nemotron 3 Super and other capable models are all there. A single API Key handles everything, and it takes 3 minutes to set up.

图片

What is OpenRouter? Which free models are there

OpenRouter is an AI model aggregation platform. You can think of it as the 'app store' of the model world. One API Key can call models from hundreds of different vendors, without registering with each one separately.

The best part is that a batch of models is completely free, not billed by token, with no monthly fee. According to OpenRouter's official Free Models page (May 2026 data), the current free models include:

First, general conversation. Owl Alpha — 1.05 million token context, native tool calling, the first choice for agent workloads

• OpenAI gpt-oss-120b — OpenAI's open-source 117B MoE model, runs on a single H100

• Z.ai GLM 4.5 Air — Zhipu's lightweight flagship, supports toggling 'thinking mode'

• DeepSeek V4 Flash — 284B MoE, 1 million token context, a speed-oriented performer

There are plenty in the coding direction too. Poolside Laguna M.1 — optimised specifically for code agents, 128K context

• Baidu CoBuddy — Baidu Qianfan's coding model, supports 65K output

• NVIDIA Nemotron 3 Super — 120B hybrid MoE, top five on coding leaderboards

For multimodal, Google Gemma 4 31B supports text and image input (256K context), and Nemotron 3 Nano Omni covers the full range of text, image, video and audio modalities.

图片

There are also lightweight models such as gpt-oss-20b, Nemotron Nano 9B and Gemma 4 26B. Altogether a dozen-plus vendors covering the three scenarios of conversation, code and multimodal.

What is the cost of free? Some models record your prompts for improvement, and requests may queue during peak times. But for everyday use it is entirely sufficient — I will cover this in detail in the hands-on tests later.

3 minutes to register and get an API Key

The registration flow is extremely simple, and no credit card is required.

Step 1: register an OpenRouter account

Go to openrouter.ai and click Sign Up in the top right. Google one-click login is supported, or you can register by email. Fill in your email, verify it, and you are done in seconds.

图片

Step 2: create an API Key

After logging in, click the avatar in the top right and select Profile → API Keys. There is a 'Create Key' button in the middle of the page — click it, give the key a name (anything, like 'Hermes'), then copy and save it.

This key is shown only once, so if you lose it you must create a new one. Store it in a password manager or a notes file.

That completes the OpenRouter part. Next, connect it to Hermes.

Hermes Web UI one-click import, 23 models ready at once

Hermes is an open-source AI agent framework with a Web UI for managing multiple model channels. Use it to connect OpenRouter and configure once to use all free models.

Step 1: open the Hermes Web UI and go to channel management

In the left menu find 'Channels' and click 'Add Channel'.

图片

Step 2: choose OpenRouter and paste in the API Key

Select OpenRouter as the channel type. Paste the API Key you copied, and leave the Base URL at its default (usually openrouter.ai/api/v1). Click confirm to save.

Step 3: verify the connection

After saving, Hermes automatically pulls OpenRouter's list of available models. Once the connection succeeds the channel status turns green, and the model count shows 23 or more depending on how many free models OpenRouter currently has.

That is the setup complete. From opening the page to having models available, 3 minutes is enough.

You can now pick a model and chat directly in Hermes's conversation interface. All free models are in the dropdown — click whichever you want.

Hands-on: what free models can actually do for chat, code and PPT

Saying they work is not enough — let us look at real results. I tested several scenarios in Hermes.

Scenario 1: everyday conversation

I chatted for a few rounds, asking a few common-sense questions and some requiring reasoning. DeepSeek V4 Flash and GLM 4.5 Air had the fastest reply times, essentially instant. Owl Alpha performed well on complex reasoning but occasionally took an extra second or two.

For everyday chat, questions and emails, free models are entirely sufficient.

图片

Scenario 2: code generation

I had it write a dynamic web page (the kind with CSS animations). Poolside Laguna M.1's code quality exceeded expectations — the generated code basically ran on the first try with attention to detail. gpt-oss-120b was also steady at coding, being an open-source model from OpenAI after all.

For simple scripts, front-end pages and data processing, free models handle everyday coding tasks fine. Complex multi-file projects may still require a paid model.

Scenario 3: PPT / skill invocation

Hermes supports a skill (skill document) feature for adding specific capabilities to models. I tested a PPT-generation skill, having the model auto-generate slide content from a topic. The result was middling but well structured, though layout details need manual adjustment.

Perfectly adequate for everyday use, and impressive in coding scenarios. Complex tasks still need a paid model. It is free — what more do you want?

A few pitfalls worth knowing

I used it for a while and hit a few snags, so let me flag them in advance.

Free models have request rate limits. At peak times (8-11 pm in China) some popular models may queue and responses slow down. If you hit a timeout, switch models — that usually solves it.

Some models need the :free suffix. When calling OpenRouter's free models in some clients, you must append :free to the model name to hit the free quota. Hermes Web UI already handles this detail for you, so no manual suffixing is needed.

Free models record prompts. OpenRouter's official note states that some free model providers may record your input for model improvement. Do not send sensitive data through free models — a paid model or local deployment is safer.

The model list changes. OpenRouter's free models are not fixed; they are adjusted up and down with partner arrangements. Check the latest list at openrouter.ai/collections/free-models every month or two.

图片

One API Key, 3 minutes of setup, and you get free access to a dozen-plus vendors' large models. For individual developers and AI enthusiasts, this is the best-value multi-model experience available right now.

Next time you want to try a new model but hesitate at the cost, spend 3 minutes running through the steps above. OpenRouter's free model list will change, but the 'one Key for all models' idea will not go out of fashion.

Sources

• OpenRouter Free Models official page: openrouter.ai/collections/free-models

• OpenRouter official docs: openrouter.ai/docs

• Hermes Web UI GitHub repository: github.com/EKKOLearnAI/hermes-web-ui

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