Kimi K2.7 Code HighSpeed is INSANE? artwork

Kimi K2.7 Code HighSpeed is INSANE?

AI News Today | Julian Goldie Podcast

June 18, 2026

Kimi K 2.7 Code High-Speed Update: Quality vs Fast vs No-Think (Real Speed Tests)Kimi K 2.
Speakers: Julian Goldie
**Julian Goldie** (0:00)
So, KimiK2.7 code has just released high speed. This is a brand new update from Kimi. And basically, this is rolling out right now. It's up to six times faster. I actually tested it, and I tested it across the non-thicking and also the other version of KimiK2.7 code, just to see how it compares. And I'll show you what we learned today and what we've got so far. So, this is basically how it works. And there's basically like three different versions of KimiK2.7 code now, right? So, there's three different gears. That's the way that I see it. It's one model, but three different ways to run it. So, before this, you know, KimiK2.7 is one coding model. What changes now is which lane you use. You've got the quality mode model, which is Kimi for coding, and that's like the full model with thinking mode on. It works like the problem out first, then rise it, and that's a careful one. That's not the new update. I mean, it came out two days ago, but it's not the new, new update. Then you have fast mode, which is the high-speed lane, which is a faster way to serve the same model, but with thinking mode on, and then you actually have thinking mode off, which is super high-speed, right?
But the only problem is like the think out loud steps are switched off. So it answers straight and answers quickly, but it's not as good. But let's see what the outputs are. So we've actually got some tests here in terms of side-by-side, what we got, using the same sort of prompt, right? So this was the prompt, which was build a 3D solar system, the sun plus all eight planets orbiting at different speeds with small labels. And you can see the outputs here. So this is like the normal sort of quality model. And then this is the fast mode with thinking mode on. And then this is no think. Now we actually compared the speeds of each response as well. So we measured those. We got 71 seconds to build this with quality and thinking mode. Then fast was 65 seconds. And then no think fast was 47 seconds. So it's almost two times faster than the quality mode. And again, like it depends what you build, but it can be up to six times faster. Having said that, if you look at this, they kind of look similar.
But the way that they're built looks slightly different. There's not a huge difference between each of them. Maybe someone who's a bit more science-minded will be able to judge on that, to be honest. But let's have a look on the next one. So the next one, we gave it a complete playable game.
Again, tested no-think versus fast versus quality.
So if we actually look at the speed, the quality responded way faster to build out, which is weird. So this took 125 seconds with no-think, which is kind of weird. So sometimes it's actually slower using the no-think, which I don't understand how that's happened, but it did. You know, we measured this directly. So these are the games that we built, as you can see right here, and they're all pretty much the same, but that was interesting too. Now all three are genuinely playable, but fast and no-think both wrapped the play field in a 3D framed border, which adds more depth. So you can see the 3D border over here. I think that's why it was a bit slower to respond, which is quite interesting.
So it's not always faster, basically. That's the whole point here. And then it really does depend on what you build. I also created a spiral galaxy here. I actually think if I look at these, fast mode with no think created the nicest version. That looks the nicest out of all of these.
If I had to pick one, I'd go with that one. So sometimes no think as well, it can reply faster. So 39 seconds versus 98 seconds for the normal mode.
But it creates something better as well. It's really, really mixed. Depends what you're building and that sort of thing. But I've not seen a 6X speed increase, unless you're using a super basic question. If you look at this, 98 seconds versus 53 seconds versus 39 seconds here. So those were three tests that we ran with this. Now you might wonder, okay, what actually happens under the hood?
What's going on here? How does it work? So the way that we set this up, you can use this however you want. But the way that I've built with this is we've got KimiCode here, and then we can switch between quality, fast and no think. So these are basically toggles that we can click between them, and then we can use the agent here. So if we say, okay, you're here, inside the chat, Kimi is thinking, and this is like normal mode. If we use fast mode, it should be a lot faster to reply, and then if we use no think, that should be even faster. So that's how you can switch. Also it applies slightly differently on no think versus fast and thinking mode. So those are like the same two replies, with thinking and thinking fast mode. So here's how it works. You got like one prompt, it's going to think first with the quality mode, then it's got standard lane, then it's going to build. With fast mode, it thinks, high-speed lane, and then it builds.

5 more minutes of transcript below

Feed this to your agent

Try it now — copy, paste, done:

curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire.

Using your own key:

curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000773196325