SGLang supports Kimi K3.1 Pro on day one, reaching up to 33.7K tokens per second per GPU on GB300

Thu, Nov 5 00:26 UTC+8

Sources1

@sgl_projectXToday 00:26 UTC+8Kimi K3.1 Pro replaces dense MLA with sparse attention (DSA). It's much more efficient and easier to serve. SGLang delivers on Day 0: · Up to 33.7K tok/s per GPU (AgentX, GB300) · 450 tok/s single user with DFlash2 + MegaKernel · NVIDIA Vera Rubin / Blackwell / Hopper, AMD, TPU, Trainium, PPUView original · Today 00:26 UTC+8

Headline by dalao.ai. The quoted post is the record; its author keeps authorship.