Moonshot's Kimi K3.1 Pro ranks fourth on Artificial Analysis index

Kimi K3.1 Pro is a post-trained version of K3 that received significantly more post-training compute. It scores 55 on the Artificial Analysis index, 11 points above K3, and sits behind Claude Opus 5.5 (58), Claude Sonnet 5.5 and Muse Wonder Full (both 56). A faster variant, K3.1 Pro Highspeed, is also available, which Moonshot says runs two to three times faster.

The largest gains are in agentic work. Moonshot reports 65.7 percent on Terminal-Bench 4.0, up from 13 percent for K3, and Artificial Analysis gives the model 1,846 Elo on GDPval-AA. On Moonshot's internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent to 8.3 percent. The model also works with Unity, Unreal Engine and Godot, and can write scripts for Blender and CAD software without MCP connectors.

Input pricing is $2 per million tokens, down from $3, and output is $10, down from $15. The parameter count stays at 2.8 trillion and the context window is one million tokens. For the open-weights release, Moonshot removed cyber-related data from both training phases, which lowered its internal xdev-bench-hard score from 90.5 to 81.0 percent.

Developers evaluating models for coding and agentic workloads now have another option at a lower per-token price than K3, though most of the headline numbers come from Moonshot or Artificial Analysis. If the open weights ship as planned, teams that want an open model near the top of the index will have a new candidate. Because cyber-related data was removed from the open release, run your own security-related tests before relying on it.