Kimi K3 goes live on Modal with custom DFlash speculative decoding for lossless speedup

AAAzzam · x · 2026-07-27

Kimi K3, the first open 3T-class model from Moonshot AI, is now available on Modal. Modal trained a custom DFlash speculator for K3's architecture, delivering faster inference with no quality loss. K3 features 2.8T parameters MoE, 1M context, and native vision.

Original post →

More from Models

Models channel →