Moonshot AI releases Kimi K3, a 2.8T-parameter open-weight MoE model

Moonshot released Kimi K3's weights and technical report: a 2.8T-parameter MoE model with native visual understanding and a 1M-token context window. Day-0 partners include Together AI, Nebius, DigitalOcean, and Perplexity; Moonshot also open-sourced FlashKDA, a CUTLASS attention kernel delivering 1.72×–2.22× prefill speedup on H20.