[Emergency Episode] Moonshot’s Kimi K3 has Arrived! China has a Frontier Model
Episode publishedPodcast: SemiAnalysis WeeklySemiAnalysis Weekly
Transcript preview
Preview only
All right, Max, we're gonna do a podcast. We're gonna talk everything about Kimi K three and maybe some other models that just came out. How are you doing?
Doing great. Looking forward to it. And thanks for having me, Jordan.
I'm not having you. Josh, thanks for having me. Okay. On the docket, let's say, Kimmy K three hot takes. Is this the third best model in the world impact on OpenAI and Thropic architecture changes, personal usage that we've had so far? What we think about their open source strategy and maybe more. All right, Max, quick hot take. Is this the third best model in the world right now?
I think the answer is a clear yes. People, people love shooting on benchmarks. I think benchmarks definitely have their problems, but I think sort of if you take a composite of all the main benchmarks and just look at the model rankings, they have been directionally correct over time. And I think if you look at that composite today, there's, like, a pretty clear top three with Fable, Soul 5.6, and now Communite three. And there's sort of just, like, always above everyone else, which includes, of course, other open source guys, like, you know, Deepsea, Conglion, whoever. But also, like, very no
Yeah. They in the limitations sections of the blog post, they said despite being a highly competitive model overall, k three nonetheless exhibits a noticeable gap in user experience compared with Fable five and GPT 5.6. My experience using this personally is that it is good. It's really slow, which is really annoying. It's motivated me to try open source harnesses for the first time, and so I feel like I'm learning more about the harnesses than I am about the models because, frankly, all these models are, like, good enough to do the basic work that I've been doing so far. I I can't really find
What do think
the strategy is for the delay between announcement of the model, the API being available and no waits yet.
Yeah. I mean, I think, to be clear, this is all just pure speculation on my part. But I think one big reason is, like, they need to give the VLM and SD LAN guys enough time to make sure they can serve, like, this model performantly because and they just, like, dropped it today. You have all this hype, but then, like, everyone else serving the models won't give me you, like, 20 token per second or something. That's probably really bad for their brand. It doesn't really sort of, like, capture. So they have this, like, incredible opportunity to get a bunch of huge PR and a bunch of adoption, and
Yeah. Makes sense. And I think, functionally, this is really interesting because just to talk about the model architecture for a second, it's 2,800,000,000,000 parameters. This does not fit in a b 200. So you need to have b 300, g b 300, or I guess m I three fifty five x in order to be able to serve this model on a single system, like a single eight way HGX server. Of course, you can do unique things where you have pipeline parallelism across multiple nodes and stuff, but that's gonna really impact performance. So I think there's a lot of recipes being cooked up and only people who have the la
Keep exploring with a free account.
A free account preserves the short preview. Upgrade to Pro for the full transcript and episode chat.