Sakana Fugu Ultra 1.1 DESTROYS Fable 5?

**Julian Goldie** (0:00)
Today, we're going to be looking at Sakana Fugu Ultra, which is a new model that's just dropped. This is from Japan. And if you look at the benchmarks here, Sakana Fugu Ultra 1.1 is actually beating Fable 5 on the benchmarks. Now, I've run several tests on it. I'll show you what they look like in a second. But yeah, basically Sakana Fugu 1.1 just dropped, and it's unlike any other AI model you've used, because you can give it one prompt, and it sends a whole team of AI experts to build it for you. I actually put that team through the hardest game builds on my benchmark, and it actually matched the best model on the planet on its very first try. So I'm going to walk you through some of the top builds so you can see it yourself. Now, if we go over to Goldy Bench, and we check out the stuff we've created, let's have a look at some of the stuff we have built with this. We've had it running for hours. Just to see what we can create with it. So this is Fugu Ultra 1.1. And you can see some examples here. So it creates some pretty nice stuff. I mean, this game is super fun to play. 3D looks pretty cool. I also compare it to Fable 5 and the same sort of tests in a second. So you can compare them side by side. So let's have a look at the next one. This is like an open world game as you can see. You might say, well, why don't you use it for business automations? I usually do, but the difference is that when you're using this for coding out games, you get a good feel for how well it can plan, how well it's good at logic, how well it can code, etc.
Also what the UI is like when it designs stuff. So it's a really good test of what works and what doesn't. This one is pretty cool. Wow, it can create some amazing stuff as you can see. Bear in mind, this is one shot. What happens with Sakana is that it uses several different AI models to put things together and then create them. For example, if we look at this right here, it's basically got a council of models and that's how it gets better outputs than for example, Fable 5 on its own.
If we have a look at this one, it's like a Skyrim style game. Fun to play, easy to use, open world, all this stuff is really good. The only downside with Sakana, and this one actually blows me away, is that it's one shot. You can't go back and forth with it because it takes like 15 minutes to generate one shot. When we're using this, all of this stuff that you're seeing right now was created first time around. We didn't have to go back and forth here and find that. But it creates amazing stuff. It's like a parachute game as you can see.
Pretty cool. I like the detail and everything like that on the game. And this is a flight simulator.
Pretty nice. Everything worked first time around as well. Let's have a look at this one. Yeah, pretty cool. All right. So how does this work?
Basically, you give Sakana Fugu a prompt, then you get the orchestrator that routes to one to three experts. And then you can use the system to get one build. So it merges all of the answers together. And the expert team is working together separately. And then these get fused inside one final answer. So if we have a look at this system, we've actually got it inside the agent operating system. If we scroll down to Sakana Fugu, we can ask the council of models questions over here. And then we can see everything inside our workspace over here. So all this sort of stuff has been built directly using Sakana Fugu and all the builds you saw before. So everything that we create with this as well, it lands in our workspace so that we can reuse it later.
Now, why do we do this? Well, the thing is, there's a one brain sealing problem. So every model you use has the same hidden limit. One brain does a whole job. So the same brain plants your physics, paints its graphics, wires its score book. And when the job gets big, that one brain can start dropping things, for example, like context. And so the graphics might land, but the controls break or the physics might work, but part of it never renders. And so we've all been there. Like a long build comes back 90% right and 10% broken. Now with Sakana Fugu 1.1, particularly the Ultra model, you stop using one brain and you route the job to a team of experts and merge what comes back. And that's Fugu 1.1. Now you also might say, okay, Fugu and multi-agent sounds a bit like it might be slow.