Tencent HY3: NEW Chinese AI (FREE + Open Source!)

Today we're going to be testing out Tencent He, which is a brand new open source project that's available on a free API, and we've been building all sorts of stuff with it this morning. I'll show you how it performs, how it goes, etc. We've already built it into our AgentOS, and you can see an example of it working right here. Now this is open source, it's just dropped from China.

I'll show you the benchmarks in a minute. If you want to use it, basically you can get it on the API for free with OpenRouter. And it's available on OpenRouter for free, as you can see right here. If you just type in HY3 free, you will see it over here.

And then you can get the API and plug in it. So that's pretty cool. It's available until July the 21st on the free API. You can also get it on Noose Research with Hermes Agent for free, and it's also available on Kilo Code for free as well.

So, open source, free, what's not to like? Well, I'll show you some examples of what I do like and what I don't like. So you can see actually here, the website that it created is not bad. Not bad at all.

We said build out a beautiful landing page. It went off and built it, and we can see the preview. And we can see our saved creations over here. So not bad, right?

Not a bad start. However, let me show you what I do and don't like about it. So you can see an example of what we built here. This is kind of like an open world game.

And every new model that drops, we test it out on GoldieBench. And then we see how it performs when we compare it to everything else. Now, if you have a look at this game, it's not bad. It just, it doesn't feel quite right.

There's something about it. I would say that's probably one of the better creations from H1-3. But you see how that light over there, it just looks a bit weird, right? It's just a bit glitchy.

And this open world game is not bad, but it's not really playable. It's not really fun to use. Let me show you another example. So you can see here, it's kind of like a Lego man that's been created on this example.

And it's an absolute giant compared to everything else. But if you look at the graphics and the details of this, it still feels very basic when we're using it. And that's something that I've actually found across pretty much everything that we tested it with. So you can see another example here.

We can like drive around a city. I would say this is probably the best creation that we got so far. It looks quite cool. It's quite fun to play.

But if you compare that side by side versus, you know, something Frontier like Fable, or you compare it versus like GLM 5.2, which we'll do in a second, I would say it's not really on the same level. Like even look at the city blocks here. They're just super basic and that sort of thing. So it's awesome that it's free, awesome that it's open source.

Is it the best model or one of the best models I've ever used? No. Would I say it's right up there? Probably not.

Like see how it's super glitchy in terms of using this and then we fly in the air. There's no detail. There's nothing going on as well. I'll show you the final example here.

So that is for this parachute game, which is quite fun, like fun background and that sort of thing. If we release a parachute, that's what it looks like. It's just not quite on the level that I would hope. And we did do a lot of back and forth on this as well.

So if we actually have a look at our HY3 tests, we've been coding and testing this out. Let's have a look here. For like two or three hours. So you can see even like two hours ago, we were testing the flight simulator and it still wasn't quite right.

So we've been going back and forth and wrestling with it for like three to four hours. It's still not getting the outputs, even when we give a lot of feedback to Fable 5 to test it out more. Now, if you're wondering, OK, how does it perform on benchmarks? Let's pull them up.

So here are the benchmarks. And you can see this is versus other models like Claude, GLM 5.2. And honestly, from what I'm testing, you can see on MCP Atlas, BrowseComp, et cetera, Claude Eval, it's kind of, well, it's beating GLM 5.2 on Claude Eval, which is like an agentic workflow. When I've tested it out myself, it doesn't seem to feel like that at all.

It doesn't seem to be on the same level. So let me show you an example of what we've built with our models using the same system. So this is the test, as you can see. If you compare that to other models, you see the quality of that versus, and this is with Fusion.

The quality of this is way, way better versus HY3. So there's a lot more detail on most of the models that we tested out with the same sort of prompts. Let's have a look on the driving game as well. So this is with Fusion.

And you can see, like, the quality is so much better than what we saw with HY3 before. This is with Hermes Mixture of Agents. This is with Fable 5. So it's nowhere near the same frontier level.

And it is open source, so I don't expect that. If we compare it versus GLM 5.1.2, I would still say that GLM 5.1.2's output is way nicer, way better, in terms of the vibe, the feel, the colors, the detail, et cetera, versus HY3. Having said that, what I do like about it is it's not bad on Hermes. So we plugged it into Hermes here.

And I think this is really what it's designed for. I don't think it's designed to be like a frontier coder or anything like that. But if we test out HY3, which we've set up the profile for here, and we say, for example, like, open up Google, it seems to work pretty well, right, for browsers. If we say forward slash learn, learn this skill, that works pretty well as well.

So it can read pages pretty quickly. It's pretty fast to use as well. And if we say, like, code out a beautiful landing page, it's not bad. Like, not bad at coding at all.

I've seen worse from paid models. So it's not bad on that perspective either. So it has pros, it has cons. I mean, one of the best things I think you can do with it is use the HY3 coder like this.

So you can basically build out a coder using the free API from OpenRouter. You can put your prompts on the left-hand side, have your previews on the right-hand side, and then you'll save stuff down here. That's really cool. Like, that is fantastic.

We also tried creating a video with it as well. So if we go back to HY3, you can see the video promo that we created up here, which is not bad. So basically what we did here is we gave HY3 access to ReMotion, and then it can code out for us. So it can create, like, full videos, full animated videos using the ReMotion skill.

And that's certainly better than a lot of open-source projects that I've created. So it's pretty nice. That video itself is good. And again, I think this is much better for agentic tasks, not for coding out frontier-level tasks, right?

So, for example, if I'm building out the agent OS, I would, every day of the week, I would much prefer to use GLM 5.2, or, for example, Claude, Claude and Fable 5, or Opus 4.8. But if you're using it agentically and you just need, like, a free, quick model that can do tasks and run tools, then I think actually HY3 is pretty good. Now, if you're wondering about the technical details of HY3, so it's a 295 billion parameter model. It's got a mixture of experts.

So it has 21 billion active parameters. If you were trying to run it locally, let's have a look. I can't see it on LM Studio yet, but I imagine that's going to drop pretty soon. Interestingly, the older models were pretty small, like only 8 gigabytes.

That's pretty small for a local model. So it'll be interesting to see if you can actually run it locally. Again, if you read the model details, you can see that it's really designed for agent capabilities. I wouldn't say it's the best open source project from China.

I think that I would pick GLM 5.2 first and then Quen 3.7. DeepSeek, I don't really tend to use so much. On the actual benchmarks, it's performing pretty well. But that's why we created GoldieBench, it's really just to test that stuff out.

So if you're not sure, then you can test it yourself or just have a look on GoldieBench and see what you think. So we've got everything that we've built with it so far here. We'll run it on all 45 tasks later today. And if it's free, you might as well check it out.

You might as well test it for yourself. So thanks so much for watching. If you want to get our full system with HY3 coder built in, and we've also got a system for Hermes where you can plug in HY3 there as well. And this is our powerful agent operating system for basically building and automating anything and plugging it all together inside one beautiful system that works amazingly well.

Then you can get that inside the AI Profitablearium. Link in the comments description or go to the AI Profitablearium.com. If you like open source projects and local models as well, we've got trainings on GLAM 5.2 and other Chinese models, for example, like Quable, that you can test outside here too. This is our community for learning and growing with AI automation.

You can ask questions inside the community. I personally answer them with video tutorials every single day. If you want to get our agent OS system, it's inside the classroom. Just go to the agent OS system right here.

And we update this daily with a video tutorial, full guide and a zip file to install it. And then every time something new and useful comes out, we actually test it and then create a video tutorial and a full step-by-step guide on how to implement it. Inside the calendar, you can jump a weekly coaching course, 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.

So feel free to get that link in the comments description or just go to the AI Profitablearium.com. Thanks for watching.