Kimi K3.1 Pro is live: $2 / $10 on the Kimi API, a 1M-token context, open weights by Nov 4

Moonshot launched Kimi K3.1 Pro today. It is on the Kimi API now at $2 per million input tokens and $10 per million output, a faster Highspeed version costs $4 and $20, and Moonshot says the full weights will be released by Nov 4.

THE PRICE

From Moonshot's API pricing page, read directly:

kimi-k3.1-pro: $2.00 input, $10.00 output, $0.20 cached input per 1M tokens. Cache writes cost $2.00 (5-minute) and $4.00 (1-hour). kimi-k3.1-pro-highspeed, which Moonshot says runs 2 to 3 times faster: $4.00 input, $20.00 output, $0.20 cached input.

That is cheaper than Kimi K3 ($3 / $15 on the same page), half of Claude Opus 5.5 ($4 / $20) and a fifth of GPT-6 Astra ($10 / $50).

WHAT MOONSHOT SAYS

A 1M-token context window, 2.8T total and 104B active parameters (mixture of experts), and multimodal input. Its tech blog calls it a post-trained version of Kimi K3 with much more post-training compute, and the first model in a new K3.1 series. It is live on Kimi.ai, Kimi Work, Kimi Code and the API. The Kimi app showed it about an hour before Moonshot's 17:15 UTC announcement.

THE BENCHMARKS ARE MOONSHOT'S

On Terminal-Bench 4.0, Moonshot reports 65.7%, just behind Claude Opus 5.5 (66.4%) and ahead of GPT-6 Astra (58.2%). It also reports 64.4 on OfficeQA Pro v2, which it calls state of the art, and a claim-level hallucination rate on its own internal test down from 23.4% for K3 to 8.3%. These are Moonshot's own runs, not independent results.

Moonshot also says it removed cyber-related training data for the open-weight release. On its internal cyber eval, the same pipeline without that removal solved 90.5% of problems; the shipped K3.1 Pro solves 81.0%, which it says remains behind frontier proprietary models.

WHAT WE DO NOT KNOW

The license for the weights, and how K3.1 Pro does on more independent boards.

UPDATE, OCT 24

Artificial Analysis now scores K3.1 Pro 55.1 on its Intelligence Index, read from its public leaderboard at 09:15 UTC. That is fourth among the models ArtificialWatch tracks, behind Claude Opus 5.5 (57.6), Claude Sonnet 5.5 (56.0) and Muse Wonder Full (55.8), and ahead of Claude Fable 5.1 (53.4) and GPT-6 Astra (52.7). It confirms the 55.1 that was circulating at launch. OpenRouter now lists moonshotai/kimi-k3.1-pro at the same $2 / $10.