Moonshot AI releases Kimi K3, 2.8T open-weights MoE model

Kimi K3 packs 2.8T parameters in a sparse MoE with native vision and a 1M-token context window, claiming 2.5x better scaling efficiency than Kimi K2. It tops Code Arena full-stack rankings, hits 89.4% pass@4 on DeepSWE, and is live on Together AI, Perplexity, AWS, Nebius, and DigitalOcean.