Back to news
researchMoonshot AI2026-10-27

Kimi K3.1 Pro scores 55 on Artificial Analysis Intelligence Index, ranking fourth

Korean media reported on Oct 27 that based on Artificial Analysis results Moonshot AI's Kimi K3.1 Pro scored 55 on the Intelligence Index, ranking fourth, ahead of GPT-6 Astra, Claude Fable 5.1 and Google Gemini 4 Argon at 53, with Meta Muse Wonder, whose API is not yet open, also included in the table. The model is a post-training upgrade of Kimi K3 from July, using a mixture-of-experts architecture with 2.8 trillion total parameters, 104 billion active parameters, 896 experts and 16 activated per token. Terminal coding is its standout area, with Kimi K3.1 Pro scoring 67 percent on Terminal-Bench 4.0 for first place against Kimi K3 at 13 percent and Claude Sonnet 5.5 at 64 percent. In Artificial Analysis's GDPval-AA evaluation it reached 1846 Elo, 312 points above Kimi K3, and Moonshot says it beat two competing models on the OfficeQA Pro v2 financial analysis benchmark. It generates about 73 tokens per second, below the evaluation average of 81, producing around 150 million tokens during evaluation against a median of 87 million. Input pricing is $2 per million tokens and output $10 per million, with Artificial Analysis estimating average cost per task at about $2.11 against roughly $5.98 for Claude Opus 5.5 at maximum reasoning. Separate reporting noted a 3D Make feature in the Kimi web and app versions that generates Blender or CAD scripts from text, sketches and reference images for online viewing and editing.

Kimi K3.1 Pro月之暗面Artificial AnalysisTerminal-Bench评测3D Make