01What is Codex HY3 free and how it works
Codex HY3 free is a way to use OpenAI's Codex coding agent with Tencent's HY3 model for free. HY3 is a new open-source model from China that costs nothing to use through the OpenRouter API. It is designed for agentic coding and reliable tool calling.
In this episode, Julian Goldie shows how to plug HY3 into Codex. You give Codex a task, and it writes files, runs commands, and reports back. Instead of paying for a model, you use the free HY3 API as the brain behind Codex.
This lets you build and test code without spending tokens.
The setup uses two parts: Codex gives you hands, and HY3 gives you the brain. Codex is free to install, and HY3 costs nothing on OpenRouter. This makes it a great option for developers who want to code more without worrying about costs.
02Main features of the HY3 and Codex setup
This setup comes with several useful features. The first is Goal Mode. This lets you give Codex a goal, and it works toward that goal automatically.
You can also see everything you have created in a workspace.
- Free API: HY3 is free on OpenRouter for about two weeks.
- Open source: HY3 is an open-source model, unlike closed models like ChatGPT.
- Agent profiles: You can use HY3 as an agent profile inside Hermes or directly with Codex.
- Model switching: You can switch between HY3 and OmniRoot in a dropdown when you get rate-limited.
- Background tasks: The agent can run in a tab while you work on other things.
The system integrates directly into your workspace. You can preview builds, save them, and come back to chat with the agent anytime. This makes it easy to manage multiple projects.
03How to set up Codex with HY3 for free
Getting started with this setup is simple. Julian shows the exact steps in this episode. You need to get an API key from OpenRouter first.
- Go to OpenRouter and get a free API key.
- Install Codex locally if you have not already.
- Ask Codex to use Tencent HY3, providing the documentation from the OpenRouter page.
- Give Codex access to your API key.
- Ask Codex to create a version where you can use HY3 for free.
Julian did this with Cloud Desktop to build it into his AgentOS system. He simply told the agent to set up HY3 with Codex, using the free API. It did not work the first time, but after quality control, it worked on the second try.
You can also do it manually by editing the config file, but Julian prefers to let an agent handle it since he is non-technical.
04How to use the Codex HY3 free setup
Once you have the setup ready, using it is straightforward. Julian shows an example where he creates a snake game as a test task. You give the agent a task, and it works on it directly in the prompt area.
You can switch between HY3 and OmniRoot in a dropdown. You can also choose between auto-approve, ask, or YOLO mode. This lets you control how much the agent does without asking you first.
In auto-approve, it runs commands on its own. In ask mode, it checks with you before taking actions. YOLO mode gives it full freedom.
You can have the task run in the background while you do other things. You also get a preview of what the agent builds. Everything you create is saved in the workspace, so you can come back and continue the conversation later.
Julian also created a separate free coder with HY3. This works similarly to Codex but runs directly with HY3. You prompt it, get a preview, and save builds to your workspace.
05Results and comparisons from the episode
Julian tested HY3 for coding and found it to be pretty decent. He built a landing page and said it looked nice, though it is not at the level of models like Fable 5. The output came back strong on agentic capabilities.
In comparisons, HY3 is not too far off from Opus 4.8. It is not at the same level, but the big difference is that Opus 4.8 costs money, while HY3 is free. This makes it a strong choice for developers on a budget.
Julian also showed that the setup works well in practice. The snake game task worked, and the landing page looked good. The main issue was the first attempt failed, but after quality control, it worked fine.
06Tips, limits, and mistakes to avoid
There are some limitations to keep in mind. The free HY3 API on OpenRouter is only available until around July 21st, so you have about two weeks to test it. After that, you will need to swap in another free API.
- Rate limits: If you get rate-limited on HY3, switch to OmniRoot.
- Quality control: The first attempt may fail; check the output and retry.
- Token savings: Use free models for bigger tasks to save tokens on paid ones.
- Open-source tweaks: Use projects like Headroom, Caveman, and Ponytail to reduce token use with Codex.
Julian says you can swap the API in the config file manually, but he prefers to let an agent do it. This saves time and avoids mistakes. Also, remember that HY3 is not as strong as paid models, so expect some differences in output quality.
07Who this setup is for and why it matters
This setup is for developers who want to code with AI without spending money on tokens. It is useful for beginners and non-technical users like Julian, who can rely on the agent to handle the setup. It also helps people who run out of tokens on paid models.
When you run out of tokens on Claude, for example, you can switch the same project to HY3 or the Codex setup. This lets you keep building without stopping. That is a big advantage for productivity.
Julian mentions that new free APIs appear every month, like Alpha earlier this year. This means you can keep using free models as they come out. This setup makes it easy to swap APIs and stay current.
For those interested, Julian offers his full AgentOS setup with free models, local models, and custom workflows inside the AirPuffer board. But even without that, you can use the steps in this episode to get started on your own.
