All products are now available to order! $45k for tinybox green with 5090s, and we have two variants of the tinybox pro v2, one with 5090s and one with RTX6000 Blackwells (that's a whopping 768 GB of RAM)
It's really nice to have a local Kimi, this is as good as the best models just 6 months ago, and you know nobody can take it away from you or stop you from finetuning so it'll never refuse your requests. With this cranked to 1000 tok/s it might even be the best experience.
🌘 Kimi-K2.7-Code, our latest coding model, is now released and open-sourced!
🔷 Improved coding & agent performance over K2.6: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on MLS Bench Lite.
🔷 Reasoning efficiency: Less overthinking, with 30% lower
A reminder to look past the hype and look at the numbers. Google is the largest owner of compute in the world. AI is not a race, it's a decentralized revolution that will take decades to play out.
This makes me not want to waste any time using it. Who knows if it's silently sandbagging me. Is tinygrad close enough to frontier LLM for it to? Just makes the model completely untrustworthy.
mythos will be bad ON PURPOSE on ai "frontier llm research" tasks, this is very very sad for the research community
also the fact that this is un purpose not visible to the user is crazy
tinygrad wants to make it as easy as possible to answer three questions about Tensor compute. What happens? Where does it happen? And when does it happen?
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.
The potential speed improvement vs JAX for large training runs is
Prices updated. For those people who placed orders, you have 5 days to land payment or they will be cancelled and you'll have to reorder at the new price. I wish we could make these machines cheaper for you, but we're actually making even lower margins with these new prices.