Cursor Open-Sources MoK Training Megakernel

Cursor Research open-sourced Mixture-of-Kittens (MoK), a deterministic mixture-of-experts training megakernel for Composer models. MoK fuses all MoE communication and computation into one kernel, achieving up to 2.37x higher throughput than the strongest public baseline. It requires NVIDIA Blackwell SM100 or SM103 GPUs and is on GitHub under Apache-2.0.