Moonshot AI has released Kimi K3.1 Pro, a 2.8-trillion-parameter reasoning model that scored 55 points on the Artificial Analysis Intelligence Index, landing fourth behind Claude Opus 5.5 (58), Claude Sonnet 5.5 (56), and Muse Wonder Full (56)3. The model gained 11 points over its predecessor Kimi K3 on the same index. Moonshot plans to release the full model weights by November 4, 2026, which would make K3.1 Pro the highest-scoring open-weights model available and narrow the open-proprietary gap on the index to 3 points4.
K3.1 Pro is a post-trained version of Kimi K3 built with significantly more post-training compute, while the parameter count stays at 2.8 trillion. The model carries a 1-million-token context window and 104 billion active parameters. It handles both text and images.
Moonshot reports that K3.1 Pro outperforms GPT6 in programming agent evaluations1. The model shows major gains in agentic tasks, coding, and 3D modeling. Moonshot also highlights enhanced financial capabilities. K3.1 Pro evaluated on Terminal-Bench 4.0, Terminal-Bench Science 0.1, DeepSWE v1.1, SWE-Marathon v1.1, and PostTrainBench.
API prices drop by 33 percent compared with K3. Moonshot is also shipping K3.1 Pro Highspeed alongside the base model.
ANALYSIS The release tightens the race at the top of the Artificial Analysis Intelligence Index, where only 3 points now separate the fourth-ranked open-weights contender from the leading proprietary model. An open-weights release on November 4 at these benchmark levels would give downstream developers access to near-frontier performance without API lock-in.
The release comes one day after Vector Wire reported that Chen Guangyu, the 17-year-old co-author of a core architectural technique behind Kimi K3, had left Moonshot AI to join Jeff Dean's Discovery Loop[1].