Chinese AI company Moonshot AI has launched Kimi K3.1 Pro, its new flagship model, positioning it close to the leading closed models from Western rivals while promising to release its weights openly. Introduced on 23 October 2026 as an upgraded version of Kimi K3, released in July, it carries about 2.8 trillion parameters in total, 104 billion active per token, and a one-million-token context window.

How it performs

Moonshot's own benchmarks place the model roughly level with Claude Opus 5.5 and GPT-6 Astra on several tests. On Terminal-Bench 4.0, which tests coding agents on realistic terminal tasks, Moonshot reports 65.7 percent, against 66.4 percent for Claude Opus 5.5. Independent measurement by Artificial Analysis scored the model 55 on its Intelligence Index, fourth overall, behind Claude Opus 5.5 at 58. Its strongest gains are on agent-style, multi-step work, including business software workflows, where its AutomationBench score rose from 58 percent to 73 percent.

Where it still lags

The picture is not uniformly strong, and Moonshot's own charts show it. On Terminal-Bench Science, K3.1 Pro scores 38.1 percent, well behind GPT-6 Astra at 68.1 percent and Opus 5.5 at 63.3 percent. On BenchCAD, which asks models to rebuild 3D objects as code, it scores 92.7 percent with tools, behind both rivals. Its measured hallucination rate also remains high on independent knowledge tests, though lower than its predecessor's.

Price and open weights

The model is priced at 2 dollars per million input tokens and 10 dollars per million output tokens, with a faster Highspeed tier at double those rates. The key date is 4 November, when Moonshot has promised to publish the full weights. Until then it is API-only, and testers will watch whether scores hold once anyone can run it.