This New Model Benches Better than Fable 5

This model is supposedly better than Fable 5. One model to command them all. Command AI. Correction. Not better than Fable 5. Better than Fable uh better than Opus 4.8. No. Where did you see Fable? Fable. I'll show you. And here we do have Fable 5. No. Somewhere we have Fable 5. Isn't it better preview? Yeah. in on terminal code bench. In terminal bench 2.1, it's better than fable 5. So technically in one place on terminal bench, uh it's uh in CTI, I don't know what that is, but on this benchmark, it's better than even Myth is preview with by getting 69 points, which nice. Um, how did they get Mythos preview? Either way, uh, if we look at it and compare it to 4.8, uh, in so many places, it's getting like a better score like in, uh, the ultra version, especially 73 versus 69. Terminal bench higher. Like pretty much it everyone is higher, right? every one every single one of these uh it's getting a higher point or like in this case equal that is really impressive but I haven't seen enough people like building things with it and being like look it's actually better I haven't seen a video from Theo about it um maybe he's cooking or maybe I haven't seen it but either way it looks really promising but you and you try to use it but It's not available yet, right? Yeah. So, I basically, if I take over right now, if I show you um the open router stuff, I was like, "Oh can we get an open router? $5 in, $30 out." Um uh 37 TPS, which actually isn't that great to be honest, but you know, 100% uptime. Same can't be said for um GitHub. um text in, image in for text out. I thought, "Oh, give this a go." So then I basically set it up with uh sweet agent cuz this is where all of these benchmarks were taken from is the mini sweet agent. Can you read that? It's quite small. is miniu agent which is a pretty barebones harness I think I think um deep suite the bench from I don't know some lab they're all using mini suite agent as well um I tried to set it up but it was just wasn't it just wasn't letting me well I say it wasn't letting me it says it's there's a provider message and it says unavailable so I'm guessing I'm guessing you have to be part of the secret crew cuz they talk a lot on this article as well. Um, they talk a lot about the um the early testers. I don't know somewhere. Here we go. Deeply grateful for our early testers. Maybe we can uh maybe we can set ourselves up for an early tester. Apply for beta test. Here we go. So, um Oh, no longer accepting. Oh, that's a shame. Yeah. So what this seems to be, I mean this is basically what their thing is is that actually it's it's one it's multi- aent workflow behind one single agent. Uh maybe even says it maybe even says it better than I could ever could. Um a full multi- aent orchestration system as a single foundation model. So whether it's their harness, there's like something on the back end that splits up the tasks and and that's all just hidden. It's like Manis agent or something going off and doing a bunch of stuff behind the scenes and then they still just return to you the one response. It's all happening behind the scenes. I don't know. Or whether it is still it's just the behavior of the model. I don't know. But that's how they've how they're able to get these crazy crazy numbers which I'm sorry. I don't know whether I believe that it's going to be better than better than Fable. Um yeah, I don't know. But yeah, I mean we've reported on these guys before and I can't actually remember what we'd spoke about, but we have reported on Sakana before. Uh but this is their new model Fugu Ultra and I'm not even sure when we're going to see it. Yeah, it's all it's all it's all snake snake oil or whatever. Don't know. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI.