Anthropic and OpenAI don't have any secret sauce.

With the release of Kimi K3 today, Chinese models are now within striking distance of the best of the US frontier models. I've seen several benchmarks that show K3 at about Opus 4.8/GPT-5.5 level, some even putting it very close to Fable. We used to think that Chinese models were at least 6 month behind the frontier models, but now it looks like maybe 2 or 3 months, if that. MoonshotAI says they'll release the ~3T weights for K3 on July 27th. Of course, to run a 3T weight model you'll need to sp