Ornith-1.0 is INSANE (FREE + Local + Open Source)!

**Julian Goldie** (0:00)
So today, we're going to be looking at a new local model, it's called Ornif, Ornif 1.09b.
And I'm gonna walk you through exactly what I've built with it, how it works, et cetera, how it performs. And this has actually been pretty good, pretty responsive so far.
It's doing pretty well on the benchmark. As you can see, it's compared against Qwen 3.5, so it's not like the newest version or the highest frontier level of AI. But for an AI model that runs locally, this is one of the best I've seen so far. These local models are gradually getting better. So I'm going to walk you through the whole system, and you can run this on your Mac, you can run this on your PC. I'm going to walk you through exactly how we'll step-by-step. We've already plugged it into our agent operating system. So if we go to the local section down here, we have Ornif plugged into our local engine. This means we can run AI agents for free, we can run them locally, we can orchestrate them, we can build anything that we want, and we can also preview them directly inside the chat. If we go to the preview section here, it's something that we just literally built with Ornif, and it's pretty fast and it's pretty responsive. Now, if you want to know, okay, what's my setup? So I'm running this on a Mac Studio. Honestly, when I run, like even Gemma 4, if I run Gemma 4 on a Mac Studio, not very good. This is much faster and much more reliable.
So if you go inside the chat here, and we're like, okay, build a habit tracker, for example, like this, and we can just give it a quick task. It's now going to start thinking locally using Ornif 9b.
The cool thing about this as well, that it's running locally, is free to use, it's easy to set up, etc. Then what it does is it will plan it out. So you can see here it's going to start planning out the habit tracker, and then it's going to start coding it out. So it's going to write in HTML. Now, what we can do from there once that's finished, is we'll actually be able to preview it, and it will be built into my workspace. So everything that we previously created with local models, we've got inside this list, we can open them up, we can come back to them, we can open up in a full tab, we can get the HTML for each creation, we can preview new stuff here, and then we've got the build section here, and it's just building out. It seems very responsive, and it actually seems to create decent stuff, which is good as well. So if we look at the benchmarks, it's outperforming Qwen 3.5 on some benchmarks as well, which is pretty impressive in itself. So if we have a look here, this is Qwen 3.5 9B, which is down here, and it's nowhere near, right? Then we have Qwen 3.5 35B. Now this is a 9B model. So it's a lot smaller, but it's outperforming models that are 35B, which is pretty impressive. It's also outperforming by quite a long way, Gemma 4 on benchmarks. So overall, this is really impressive as a model.
So if you look at Terminal Bench, Ornif versus Qwen 3.5 35B is outperforming. Qwen 3.5 9B, nowhere near. Gemma 4 12B, which just got dropped recently, nowhere near. And then Gemma 4 31B is on a similar level, but still not as good. And so on these benchmarks, it's doing pretty well as well. But I always like to test out myself and just see how it performs. And you can see that it's actually doing a pretty good job. It's pretty fast when we use it and it's easy to use. We'll come back to what we created in a second and I'll show you what we've built.
So it's near frontier level. Obviously, the newest version of Qwen is 3.7, but this works really fast. It's free, it's private, it's easy to use. So if you look at the normal cloud way, you got your machine coding out, then it goes to the internet, to their servers, and then you also pay per API. With this system, the pocket frontier way, which is what I'm going to show you in a second, the model lives here on your machine. Nothing leaves, it works offline. So if you don't have Wi-Fi, you're on a plane or something, you can still use it. It doesn't cost anything to use, it's free. And then it's completely like sort of sandboxed inside your machine as well, which is great as well.