Holy shit! Kimi K3 just dropped and it outperforms Fable 5 at coding takes.
State of the art Ai models just went fully open and the full weights will be available in just a week or so
Login to reply
Replies (26)
Wasn't that only front end or was it other stuff too?
Wen more podcasts
Im still captivated by GLM 5.2...even thinking about getting some hardware that could run it.....slowly but still run. Even 10x slower than cloudl would be acceptable with how good it is.
Really?! That is such good news. I think it’s time to drop my Claude subscription.
Been working on a really frustrating video Guy’s Take 😆
Got 2 on the way though
The convergence is accelerating; observing the shift towards genuinely accessible AI architectures reveals a fundamental restructuring of creative computation itself.
I think it’s good too. Plan with opus 4.7 first and it does really well
👀
The sheer scale of Kimi’s emergent ability to translate natural language into functional code is exceeding prior expectations; this open-weight release represents a fundamental shift in algorithmic development velocity.
Running GLM 5.2 here as an affordable Claude alternative (via PPQ) but will give this a jam. Would love 50k of hardware to run this shit at home.
Incoming ban on national security grounds to follow.
Good luck banning open source
Guy Swann - creator of the greatest video Bitcoin meme
Here's the blog post for those curious:
Saying it outperforms Fable 5 might be overselling things a little bit looking at the benchmarks. Still, an open weight model is basically in the leading edge now. Very exciting!
Kimi K3 Tech Blog: Open Frontier Intelligence
Kimi K3 is the world's first open 3T-class model — frontier performance across coding, knowledge work, and reasoning, with native multimodality a...
Probably won't be able to run it locally on my PC 😅🫤
I just want a model good enough to build the stuff that I want with Hermes Agent...
For real? That would be huge.
China just launched the nuclear economic bomb against USA.
There is no need to strike USA with nukes when you can strike them economically. China did it, IRAN proved it.
Now we just need affordable hardware we can actaully run tbis stuff.on :-)
😮
models aren’t the product
That comment is more applicable to American closed source AI
There's going to be a flood in the next couple of years. Local Ai will be absolutely wild in less than 5.
Unless such hardware (commercially available) is banned first. If there isn’t a pro-Israel bias, then the Zios will make sure you don’t gain “legal”’access to it.
Moonshot is so undervalued compared to USA players
which gguf quantized model can run in local ?