vLLM adds day-0 support for Moonshot’s 2.8T-parameter Kimi K3

woosuk_k · x · 2026-07-28

vLLM announced day-0 support for Moonshot AI’s Kimi K3, saying the model can be served as soon as the weights are public.

The post highlights K3’s scale and the supporting engineering work around prefix caching, DSpark training, and kernel optimizations. K3 is described as a 2.8T-parameter MoE model with a 1M-token context window, native multimodal understanding, and Kimi Delta Attention to make long-context inference more affordable.

Related event: vLLM Announces Day-0 Inference Support for 2.8T Parameter Kimi K3(10 posts)→

Original post →

More from Infra

Infra channel →