China’s Tencent Hy3 is FREE & Opensource!

There's a brand new free open-source AI model from China called Tencent Hanyuan HY3. It's 295 billion parameters mixture of experts model. It's open source and that means you can host it locally if you have a good setup. If not, you can actually run it for free in two different ways.

I'll show you how in a second. And on benchmarks, this is actually outperforming on some benchmarks this week. We can see for example on this benchmark, Hanyuan is outperforming both DeepSeek and Quen 3.7 Max. I'll show you my own tests in a minute and just to confirm, this is open source.

As you can see, it's hosted on Hugging Face. One of the most important things about this is that you can actually run it for free. If you're wondering how, basically what you can do, as we've done over here, is you can create a coder like this. We've already got it inside our agent operating system.

And from here, what you can do is you can use the free API that lives on Opta. And then you can plug that into a coder like this so that we can say, for example, build our beautiful landing page. It goes off and codes it. We can preview it over here.

And then we can save all of our builds over here, right? And this works in real time. And if you're wondering how to get access to that, you can actually go over to OpenRooter to get the free API. And then the actual agent operating system, I mean, you could build your own or you could get ours from the AIProfit Boarding.

Link in the comments description or go to the AIProfitBoarding.com. So it's literally just dropped. It is free as an API. It's only free until July the 21st.

So bear that in mind. It's only available for free for two weeks. So you might as well make the most out of it. And also something to note here is that you can also plug this into Hermes.

So for example, we could use it inside the Hermes Astros or the outreach tool that we've generated for lead generation over here. We could also use it inside Hermes Oracle. Because it is now available on Noose Portal for free, as you can see right here. So it'll be available on Noose Portal for two weeks, which means you can plug it into Hermes and it is an agentic model.

So this is pretty impressive because we can code out for free on the API. It's available open source. And also you can use it inside Hermes and it can actually build out decent stuff. Now, you might be at this point wondering, OK, is it better than GLM 5.2?

How does it perform versus, for example, Opus 4.8? Honestly, at that point, I would say it's not as good. So I just want to be 100% honest for you. I'm not going to BS you or anything like that.

Like, HY3, if we have a look on the benchmarks here, it can create some OK stuff. But it's never going to outperform Fable 5. It's kind of up there with Opus on some tests, but not really. And versus GLM 5.2, I would still say that GLM 5.2 is crushing on this.

So if we have a look, for example, we've got a creation from here. This is like a Doom-style game. And we created this with GLM 5.2. It looks all right, right?

It does the job. We can create stuff. It's pretty useful. And then if we compare this versus Tencent, this is probably the better option of what we've created over here.

But still, the controls are a little bit weird. So actually, I would say that creation from HY3 is actually better than GLM 5.2. However, if we have a look at something like the Dragon Realm, which is an open-source game here, where this is a creation with GLM 5.2. So it's a fully open-source game.

It looks really cool. Nice graphics, fully open world, et cetera. If we compare that versus the creation from Tencent HY3, which we've got right here, it's just not quite as nice, not quite as smooth, et cetera. So for me personally, if I had a choice between these, and we'll check the flight simulator as well.

So this is the flight simulator with GLM 5.2. If you compare that versus free, it starts off okay, but you see how it's so like buggy here. And then if we actually try and go up in the air and fly around, like that's when you realize, okay, then the physics on this sort of stuff is not great, not great. Here's another example.

So this is GLM 5.2, created this Neon City game, which is pretty fun, pretty interesting to play, quite cool. And we can like drive around the streets and that sort of thing. But if we compare that to HY3 from Tencent, just not quite as nice. So I would say if you had a choice of models, obviously Fable 5 is right up there.

I would say Opus 4.8 is just below that. Then you've got GLM 5.2 below Opus 4.8. It's still kind of comparable, but then in reality, especially if you look at this parachute game, like HY3 is nowhere near, nowhere near Opus 4.8. So it is worth testing out.

I mean, the fact is that you're not going to get Claude for free on an API. You're not going to get GLM 5.2 for free on an API, but you can get GLM 5.2 for free on ZAI, just inside the chat. But you can get HY3 for two weeks on API. So that's pretty good.

And also when we tested it here, so you can see we tested inside our agent operating system and we said, code out a beautiful landing page. And it actually created out like a quite a nice website, as you can see. So it can build out unique projects. It's pretty useful for that.

We could use it inside our workflows. I think one of the best things you could do with this is just use it for sub-agents. So you could go to your manage section over here, then you can switch to models. And then for auxiliary tasks, for delegating like smaller, heavier tasks that require a lot of tokens, but aren't requiring like frontier level knowledge or intelligence, then that's where you could use something like Hanyuan.

And also the fact that it's open source, I mean, everything coming out of China right now is all open source, which is pretty wild in itself. So using it inside Hermes, using it inside OpenClient, using it even like, for example, you could use the API and you could plug into free cloud code with the API on OpenRouter. And that could be another way that you'd free cloud code. You could also do the same with Codex.

We've actually done that with OmniRoot already, but you could do that with Codex. Those could be some super useful ways that you get the most out of this model. Now, if you actually look at the benchmarks here and the report from Tencent HY, you can see that they've really focused on agentic capabilities. So I think that's why they've given a free trial for two weeks to News Research, because it's designed for reasoning, agentic, long context tasks.

And it says it's competitive with much larger flagship models. Would I agree with that? Probably not from my own personal tests. I think it's always good to test this stuff out.

That's why we actually created GoldieBench. Here's one thing that I did find impressive, actually. So we actually gave Tencent HY3 access to ReMotion that can create videos. And then we tested creating a video with HY3 as well.

And you can see here that I actually created this fully animated video, which you can see right here. Didn't prompt it at all, didn't train it, but it took my context, understood what we were writing about, and then it could create a full video like so. Right, so it could create a fully animated video. Honestly, if I asked my team of editors to do the same, I think that they would probably struggle to create the same thing in the same amount of time.

So at that sort of stuff, it was actually pretty good, to be fair. That was probably one of the most impressive tests. But again, if you had a look at Fable 5 and what it can do with the same sort of test, it's going to absolutely crush it. It's going to crush it.

And I only say that because, you know, some people watch this and say, why would you compare HY3 versus Fable 5? A lot of people will ask me that inside the comments, and that's why I said it. So free local model, can't argue with that. You can build some cool stuff out of it.

You can use it inside Hermes. It is open source as well. Do I think it's going to replace anything soon? Probably not.

But it's an interesting sight to see that HY3 is open source. You got Kimi, K2.7, and GLM 5.2 all dropping from China. So they're always releasing these new updates. I think probably the update that you want to look out for the most is GLM 5.3.

I would expect that to come out within the next few weeks. So thanks for watching. If you want to get my full agent operating system and a community of people who learn, grow, and support each other on this journey, then you get that inside the iProfit body. Inside the community, you can ask questions, get help, and support.

I personally answer all the questions decided with a video tutorial each day. And then inside the classroom, you can get access to all of our best trainings. And you can go from complete beginner to expert with our training over here. And also, if you want to get more information on how to use the free API, how to set up with Open Router, et cetera, we've got a full guide and step-by-step guide over there.

And then we've got new guides and trainings based on what actually works. Every time we test something out, then we create a video tutorial and a step-by-step guide. And if you want to get our agent operating system that we talked about before, you can get that video tutorial with the last update date here. And you get the zip file to install it.

Plus, we do four weekly coaching calls where you can get help and support in real time. Inside the map, you can meet people in your local area, building with AI agents. And that's all available inside the iProfit body. Link in the comments description, or just go to the iProfitbody.com.

Thanks for watching.