Moonshot's Kimi K3.1 Pro scores 55 on AA, weights due Nov 4

Moonshot's Kimi K3.1 Pro landed on the company's API this week, with full open weights promised by November 4, 2026. Simon Willison reported that Artificial Analysis scored the model at 55, ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon at 53, trailing only Claude Opus 5.5 at 58. Two of the researchers on our radar flagged the post the same day.

The pricing is the sharper detail. Input runs $2.00 per million tokens and output $10 per million, matching GPT-6.1 Sol and roughly half what Claude Opus 5.5 charges. A "Highspeed" variant runs "2-3x faster, at twice the price."

"The first time a (soon-to-be) open weight model has scored this highly against the frontier," Willison wrote. He tested the model across three reasoning levels on his usual SVG-pelican prompt, calling the top result "pretty solid" while noting "Both legs are on the same side of the frame." Moonshot's own benchmarks position Kimi "approximately level with Opus 5.5 and GPT-6 Astra."