Moonshot releases Kimi K3.1 Pro
On 23 October 2026, by Moonshot AI's own post on X of that day, the company made Kimi K3.1 Pro and a faster Kimi K3.1 Pro Highspeed available on Kimi.ai, Kimi Work, Kimi Code and its API. Moonshot calls it its most capable open-weight model of the 3-trillion-parameter class, with a one-million-token context, and says the full weights will be released by 4 November; they were not yet on Hugging Face on 28 October.
Why it matters
K3.1 Pro is a post-trained version of Kimi K3 with, Moonshot says, substantially more post-training compute, and the independent measurement points the same way: Artificial Analysis puts it fourth on its Intelligence Index with 55, 11 points above Kimi K3 (44), and says it will be the highest-scoring open-weights model once the weights are out, which is a condition, not a fact on the day. First-party prices on Artificial Analysis's page are $2 and $10 per million input and output tokens. Moonshot's own results, such as 65.7% on Terminal-Bench 4.0 against 66.4% for Claude Opus 5.5, are the company's claims.
Date. Moonshot's blog has no date. The post on Moonshot's official X account is shown as 20:15 on 23 October 2026 in Kyiv time, which is 17:15 UTC; it says both models are "now live" on Kimi.ai, Kimi Work, Kimi Code and the API and "Open weights by Nov 4, 2026". The same day is given by Artificial Analysis's article (dated 23 October), OpenRouter's page ("Released Oct 23, 2026") and OpenRouter's model listing (created on 23 October 2026 at 17:56 UTC); the record is dated by all of them, and the date note says so. What Moonshot says. Kimi K3.1 Pro is the first model of a new K3.1 series and "our most capable 3T-class open-weight model to date", with a one-million-token context; it is a post-trained version of Kimi K3; Highspeed gives 2 to 3 times the speed; both are available "today" on Kimi.ai, Kimi Work, Kimi Code and the API; "the full model weights will be released by Nov 4"; 3D Make is live on kimi.ai. Its own results: Terminal-Bench 4.0 65.7% against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5; the internal Kimi Code Bench 3.0 Long-horizon 61.2% against 55.95% for K3; BenchCAD with tools 92.7%. What Artificial Analysis says. Size 2.8 trillion parameters in total and 104 billion active; context 1 million tokens; text and image input; "weights currently not published"; price $2, $10 and $0.20 per million input, output and cache-hit tokens, 33% lower than Kimi K3; Intelligence Index 55; GDPval-AA 1,846 Elo; in its own run Terminal-Bench 4.0 67% against 13% for K3. What the record does not claim. It does not call the model open-weight on 23 October: the weights are promised by 4 November. "Will be the leading open-weights model" is Artificial Analysis's conditional statement. Moonshot's results are its own.