New Chinese AI Model Is INSANE! (FREE & Open Source)

**Julian Goldie** (0:00)
There's a brand new free Chinese model that's local and open source, plus you can get it for free on API2. There is an absolute powerhouse when it comes to benchmarks. You can see, for example, NAX is actually beating Claude on the benchmarks right here. Now, this is Claude Opus 4.7, but again, it's a free API you can use, and it's also beating Gpt 5.5 on some benchmarks as well, as you can see here. So, for example, on SWE Bench, it's scoring higher than ChatGPT, and it's an agentic model designed to be used for AI agents. Now, you can see here it's available for free, and you can get it on Hugging Face, which is pretty amazing.
So, you can see the open weights right here, and this has a context window of 260k tokens. It's free to use, and it's pretty powerful for coding. This is what it's designed for, coding and agentic features. It's also good at tool calling, function calling, which is perfect for agentic use. Now, the other thing to note here as well is actually based on Qwen 3.5's architecture, that's how they built it, and we can get it for free also on OpenRouter, as you can see right here. In fact, the number of tokens being used per day is increasing a lot because more and more people are switching to this. So, if we take a look at this, for example, this can be used for free inside Kilocode, Hermes Agent, Claude Code, OpenClaude, Pi, all your favorite AI agents. In fact, we've already plugged in to Hermes Agent, as you can see right here, we were double checking it working and it actually works, which is pretty powerful stuff. Now, the way you can configure this inside Hermes is you can go to Manage, then Models in your Dashboard, and then from here, just change this over to OpenRouter and then To. Now, there's also, and this is pretty useful, there's a new profile section here inside Hermes Agent where you can actually plug it in to an AI agent, for example, a separate agent with the model specifically for N2. So that's what we've done with this model right here. And then when we go to chat with our AI agents, we can select this one and start talking to it straight away. Now, you can also use this with Claude Code as well. So the way you can use this is you can use something called free Claude Code. There's an open source project. Plug free Claude Code into N2 Pro, and then you can build with it for free as well. So if you want to see some stuff that we built, for example, you can see some examples of stuff we've built with N2 here, like a to-do list app, some cool animations, some interesting stuff right here. And we built this all for free using N2. We can even control it with our voice and then get it to build something out, which is amazing when you think about it. It's also interesting to see this is a Chinese AI model, and it's already overtaken DeepSeek. So if you compare this to DeepSeek V4 Pro, which is a paid API. Bear in mind, like these are all paid APIs, right? Claude Opus 4.7, Gpt 5.5, DeepSeek V4 Pro, paid APIs you would normally pay for. But if you're using next N2 Pro, well, that's better on benchmarks versus a lot of these APIs and it's free to use. Now, a couple of things that I'll say right here is it's pretty good at deep search. So you can see, for example, here was a question that was tested with N2. They can do a lot of deep research. You can also use it as a terminal agent. You can use it with OpenClaude. Here's an example. Web building, so building out stuff. And then also maths reasoning too. And there's also two different models. You got N2 Pro and N2 Mini.
So N2 Pro is a lot more powerful, as you can see. But even N2 Mini isn't too bad. And obviously that will be faster to respond as well. Now we can actually compare these side by side. So we've got a different profile for Hermes here versus on N2. Let's try these out and see which one responds the fastest. And so we've got both of those working side by side, as you can see here. So this is another model that we're using. And this is using N2. And you can see it actually replies faster than using the normal profile that we've got for Hermes. We've had to buy a lot more. Now, if we actually go to the Manage section here, and we check what model the other profile was on, that was actually running with Gemma 4