Benchmarks chosen by Moonshot AI. Kimi K3.1 Pro scores are Moonshot’s own runs.

Source: Moonshot AI, Kimi K3.1 Pro blog, Oct. 23, 2026, and Artificial Analysis. Moonshot chose the benchmarks and ran Kimi K3.1 Pro itself, mostly on the Claude Code harness; rival scores mix Moonshot runs, system cards, vendor blogs and public leaderboards. All models at maximum reasoning effort; Moonshot says Claude results may include fallbacks and GPT results may include cyber safeguards. Bars run 0 to 100. Comparison: R&D World.