Moonshot Releases Kimi K3.1 Pro, Ranking Fourth Globally; Open Weights Due by 4 November
The 2.8T-parameter MoE model scores 55 on the Artificial Analysis Intelligence Index and leads Terminal-Bench 4.0 with 65.7%.
Moonshot AI released Kimi K3.1 Pro, a post-trained version of its Kimi K3 mixture-of-experts model with 2.8 trillion total and 104 billion active parameters, a 1 million token context window, and text-and-image multimodal input. On the Artificial Analysis Intelligence Index the model scored 55, placing it fourth globally behind Claude Opus 5.5, Sonnet 5.5 and Muse Wonder Full, only a few points behind the leading proprietary frontier models.
The company highlighted strong agentic and financial-workflow results, including 65.7% on Terminal-Bench 4.0, the highest among evaluated models, and state-of-the-art performance on OfficeQA Pro v2 for fiscal analysis of real financial documents. Moonshot reported reducing claim-level hallucination from 23.4% to 8.3% in internal stress tests, and disclosed that hacking-related data was removed during mid-training to limit cyber misuse. Open weights are scheduled to follow by 4 November 2026, with the licence still to be announced.