Kimi K3.1 Pro

First model of the K3.1 series and Moonshot's most capable: a 3T-class MoE post-trained from Kimi K3 with dramatically scaled joint RL across hundreds of domains, a 1M-token context, always-on thinking with max reasoning effort by default, image and video vision input, and automatic 5m/1h context caching at $2/$10 per 1M tokens; a kimi-k3.1-pro-highspeed variant ships 2-3x speed, and full open weights are pledged by Nov 4, 2026.

curl -s https://modelregistry.tirup.in/api/cli?model=kimi-k3-1-pro