Overview
Open-source LLM serving engine widely used for production inference; day-0 partner for multiple moonshot-ai Kimi releases including planned kimi-k3 weight drop.
Recent Developments
- 2026-07-22: Preview of production-scale kimi-k3 support — KDA prefix caching, MXFP4 MoE, NVIDIA/AMD paths (2026-07-26-vllm-kimi-k3-preview, 2026-07-26-kimi-k3-open-weights-july-27-release)