Oct 23, 2026 · found Nov 8, 2026, 9:10 PM PT

Moonshot publishes Kimi K3.1 Pro weights; its Artificial Analysis rank fell under index v5

Verdict: Vendor-reported · Model releases · Benchmarks

Weights are on Hugging Face, but coding numbers are Moonshot's own, and the launch-day rank of fourth was later cut to ninth when Artificial Analysis changed its index.

Moonshot released Kimi K3.1 Pro on 23 October and has published the weights on Hugging Face.

Claim: Terminal-Bench 4.0 65.7%, against Claude Opus 5.5 at 66.4%, from a 2.8T-parameter model with 104B active. Moonshot
Evidence: Artificial Analysis measured 67% on Terminal-Bench 4.0 and an index score of 55, fourth place at launch. The weights are on Hugging Face under a custom license. Artificial Analysis, Hugging Face
Caveats: Moonshot ran its own model in the Claude Code harness and took rival scores from published reports, so the comparison is not like for like. Several benchmarks are internal, the license terms were not read, and the claim of restricted cyber training data is unverified. Moonshot
Why it matters: After Artificial Analysis moved to Index v5 (about 60% private questions) the model dropped to ninth. Rankings move with the benchmark. Gizmodo

Sources: Moonshot, Kimi K3.1 Pro (primary) · Hugging Face model page (primary) · Artificial Analysis launch benchmarks (independent) · Gizmodo on Index v5 (press)

Verdict history

Nov 8, 2026, 9:10 PM PT: Vendor-reported. Primary and model page read, weights confirmed to exist. Release date taken from Artificial Analysis. Model page gives no date and the license terms were not read.