Codex + HY3 is Insane (FREE!)

Today, I'm going to show you a new way to use Codex for free using a combination of Codex and then also we've got Tencent HY3 that only just dropped about 24 hours ago. So this is a brand new model from China. It's open source, and it's available for free on the API. Also, the other thing I would say here is that this API is only available for two weeks.

You might as well make the most out of it today, which is what I'm showing you. So essentially, we've plugged in HY3 from Tencent into Codex. So you can get a free API from Open Router, and then you can easily get that API and build whatever you want for it, right? Now, we've created this custom version where we can use it with Goal Mode.

We can see everything that we've created. We can check out the workspace. So if I show you an example of this, if we go to the workspace over here, and then we see what we've created, we can see everything that we've built, and we can preview everything that we've created previously using this powerful system. And the great thing about this is the HY3 is actually pretty decent when it comes to code.

It's not like Fable 5 level, but it's a free API. And also, the next time something else comes out that's free, you could just swap the APIs and change them around. So the way that you can do this is you get your API key from Open Router, as you can see here, and then if you've already got Codex installed locally, you can just ask Codex, listen, use Tencent HY3, give it the documentation from this page for OpenAI, give it access to your API key, and then say, give me a version where I can use HY3 for free with Codex. That's basically what we did.

We actually did it with Cloud Desktop instead, because we wanted to build it into our AgentOS, and it was really simple and easy. Literally, what I said was, can you set up HY3 for free with Codex, see the free API on Open Router, put it inside the AgentOS, took the information here, as you can see, found the free API, which was pretty nice, installed Codex, tested it out. It actually didn't work first time around, so I just quality controlled it, but then second time around, ready to go, right? And it's actually working.

We can see the preview here. We can see everything we've created, and we can come back and chat with it whenever we want. Pretty amazing stuff. We also, for example, created a free coder with HY3 here too.

So it's kind of similar to Codex, but it just runs directly with HY3, no Codex involved. And then we can prompt it over here. We get the preview over here. We can save builds to our workspace.

We see all the conversation history as well. And the great thing as well is like, it integrates directly into our workspace. So let's say, for example, using Claude, you run out of tokens. You're like, what should I do?

Should I stop building? It's like, no, no, no, no, no. Just take the same project and plug it into a HY3 coder or plug it into Codex instead so that you don't stop. And that's the great thing about this.

Also, what I would say here is that if you look at the system, it really runs down to like two parts. You got access to hands, right? So OpenAI's coding agent, you give it a task. It writes the files, runs the commands and reports back.

And the agent itself is free to install. And then you've got HY3, which is a brain. So Tencent's new 295 billion mixture of experts model built for exactly this. It's designed for agentic coding and reliable tool calling.

So on OpenRouter, the Tencent forward slash HY3 free version, costs nothing to use. And you can use it until I think it's July the 21st. So it gives you about two weeks to test it out. And then all you do is you swap it inside the config file.

Now you could do that manually. For me, I don't personally like to do that myself just because I'm non-technical and I can easily just give it to Claude and then literally in two rounds, we have it ready to go and implemented. The other great thing about HY3 from Tencent is that you can actually go to Hermes here and you can have a free installation, a free agent profile with Hermes too. And that way you can use it for free in two ways.

You can use it as a agent profile inside Hermes and also inside Codex. And then you can save and build everything you want. I haven't tested out with goal mode as well, but that will be pretty interesting to use. And also if you run out of tokens or if you get rate limited on HY3, you can also use another free API that plugs directly into it called OmniRoot.

Now OmniRoot is another free API that actually runs through like 90 free providers, but either way you can use Codex for free. We've also plugged it into Claude Code. So we've used OmniRoot with Claude Code and that way we can use the agent harness and the power of Codex and free Claude Code, but with free APIs and that sort of thing, which is pretty cool. So pretty powerful stuff.

If you're wondering what's going on in the background, so literally you give it a task, the profile loads with HY3, it runs through OpenRooter and the free API, three acts, and then you get the projects back in your workspace. So if we go inside here, for example, and we give it a task and we can actually switch between OmniRoot and HY3 in the dropdown, and we can switch between auto-approve, ask or YOLO mode. And so if we ask it a task here, okay, just build out a snake game, classic sort of task, it will start working directly here as well. And also we can have that tab running in the background whilst we go off and do something else.

So it's basically like one of the world's most powerful agentic harnesses with a free brain plugged inside. And we actually tested out for like being on a landing page. It looks pretty nice. Again, it's not going to be Fable 5 level, but it looks pretty nice and it was easy to set up as well.

If you're wondering how does HY3 perform, by the way, we've got the output back there. So this is HY3 and it's pretty strong on agentic capabilities. Now, if you look at the comparisons here, it's not too far off Opus 4.8. Obviously it's not at the same level, but the difference is the Opus 4.8 you have to pay for.

Whereas for example, HY3, you get it for free, right? And that's quite a big benefit. Also, HY3 is open. It's an open source model.

Whereas for example, Chachipity is closed. Now, again, if you get rate limited on HY3, no problem. You can switch to Omniroot as well. So you can switch between them.

And this way as well, if you're using an agentic operating system like we are, then you use less tokens, or you can just delegate like the bigger tasks to smaller models that are free. And that way you don't have to worry about tokens and you can minimize the amount that you use as well. You can also, for example, plug into Codex, open source projects like Headroom, Caveman, and Ponytail. They're all ways to reduce the amount of tokens that Codex uses so that you can code for more with free models.

And so if you look at the old way, it's like, you know, every task was metered. Codex runs usually on paid models only. You kind of have to ration what you do because you don't want to run out of tokens. Free models particularly usually run out on the loop and testing a new model means adding new keys and everything else, right?

With this new setup, you have one profile. You can switch between these two free models and anytime you get rate limited, you can switch between them. Plus everything you create, you can preview inside the chat and also you have the workspace for everything you've created over here. Also, you might say, okay, well, what happens when H3 runs out?

Well, then you could use OmniRoot or there'll be another free API by that point. If you've seen anything this year, you know, for example, our alpha was another free API this year. So every single month there seems to be something new free API you can test. So if you want to get my full setup with the AgentOS, we've got free code plugged in there.

We also have local models. We have OmniRoot, HY3 Coder, et cetera, and Codex. So you can use these models for free. You can get that inside the AirPuffer boardroom.

Link in the comments description or go to the airpufferboard.com. We also have a memory system plugged in here and loads of cool custom workflows for Hermes so that you can use, for example, the Hermes Oracle or Hermes Astros or the Hermes Outreach tool as you can see right here. So that's all available inside the AirPuffer board Link in the comments description or go to the airpufferboard.com. Inside the community, you can ask questions and I personally answer them with video tutorials every single day.

Inside the classroom, you can get access to all of my best trainings. If you're a complete beginner, you can go from beginner to expert over here. And if you want our new daily updates, you can get them over here, as you can see, with new daily tutorials and our full AgentOS system. We also update the AgentOS daily and you can get the zip file to install over there.

Inside the calendar, you can drop a week of coaching calls, get help and support in real time. Inside the map, you can meet people in your local area who are building with AI agents like you. And that's all available inside the AirPuffer board. Link in the comments description or go to the airpufferboard.com.

Thanks for watching. Bye-bye.