Moonshot AI announced Kimi K3.1 Pro on October 23; its launch post lists Kimi products as well as API access. It is not open weights yet, but Moonshot's announcement says: “Full model weights will be released by Nov 4, 2026.”

Kimi's own API homepage lists standard pricing at $2.00 per million input tokens and $10 per million output tokens, checked November 1. A “Highspeed” variant is 2-3x faster at twice the price. Moonshot's post gives a 1-million-token context window, and the API quickstart lists low, high and max reasoning levels.

Benchmarks

Our original October 27 report used an older Intelligence Index snapshot. AA introduced v5 that day; its current leaderboard lists K3.1 Pro at 46, checked November 1. Use the current Benchmarks page for a consistent index version rather than comparing that 46 directly with the older 55.

Moonshot's own post reports 65.7% on Terminal-Bench 4.0, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra, plus results on its own Kimi Code Bench and Kimi Hallu Bench suites. These are vendor-reported figures we have not reproduced, and they come from a different Terminal-Bench run than the one behind our Benchmarks table, so do not set them against the Terminal-Bench figures there.

What is still open

Moonshot's post names no license. The weights date is a company promise, not a leak, so it carries no rumor score, but the license is the part that matters to self-hosters. K3.1 Pro now has a Model Guide entry as well as a Benchmarks row.

Update, November 1: Kimi's November 4 weights promise is still pending. Mistral's October deadline has elapsed; our verification of its downloadable artifact is incomplete.