Moonshot AI Releases Kimi K3.1 Pro With 1M-Token Context, Plans Open Weights by Nov. 4

K3.1 Pro is built on Kimi K3 and adds post-training compute, with joint reinforcement learning across hundreds of downstream tasks. The Highspeed version is claimed to produce output at two to three times the speed of the standard version.

In evaluations published by Moonshot AI, K3.1 Pro scored 65.7% on Terminal-Bench 4.0. In an internal factual-accuracy stress test, the hallucination rate for individual statements fell to 8.3% from 23.4% in K3. Artificial Analysis, an independent evaluator cited in the report, gave K3.1 Pro a score of 55 on its intelligence index, ranking it fourth and 11 points above K3. It placed behind Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full.

The model is priced at $2 per million input tokens, $10 per million output tokens and $0.20 per million cached tokens. Moonshot AI said all three prices are 33% lower than those for K3.

The company also said it removed relevant data from its mid-training corpus and reinforcement-learning environment to reduce the risk of cybersecurity abuse after the weights are opened, while retaining general coding tasks. Kimi's enterprise edition will not enable the new model by default; organization administrators must turn it on.