Kimi K3 Leads Multiple Agent Benchmarks Over GPT-5 Variants
Source Posts on X
Prediction came true
Super good
Posts from X
2 postsSuper good
Prediction came true
Btw, Moonshot will likely have a Mythos tier model before DeepSeek 3T total parameters, 70B-90B active, they have lots of high-quality tokens, already ahead in multimodal understanding Throw in some Attention Residuals and Kimi Linear It might not be as cheap as DeepSeek, but it will be much better and more efficient