The company says it has a 1-million-token context window, meaning it can handle very long inputs, and that a faster variant, Kimi K3.1 Pro Highspeed, runs 2 to 3 times as fast.
Both are available on Kimi.ai, the Kimi mobile apps, Kimi Work, Kimi Code and the Kimi API.
Moonshot plans to release the model's weights, the trained parameters that let others run it, under an open license on November 4, 2026.
Separately, the launch has drawn scrutiny after an X user, Elshayib, posted screenshots he described as instructions for a Kimi influencer campaign telling creators not to use a paid-partnership label.
Pekingnology reports the document has not been independently authenticated.
According to Moonshot and the testing firm Artificial Analysis, the model leads that firm's Intelligence Index, ahead of Xiaomi's MiMo-V2.6-Pro, DeepSeek V4.1-Flash-1017, Meta's Muse Wonder and OpenAI's GPT-6 Astra.
Moonshot lists API pricing at $2 per million input tokens and $10 per million output tokens, about half the rates of Claude Opus 5.5 and Muse Wonder, though DeepSeek and Xiaomi remain cheaper.
Kimi is reported to do well on multistep work such as coding, debugging, tool use, financial analysis and security tasks, with a 64% aggregate on Artificial Analysis's Cyber Index.
On Moonshot's internal Kimi Code Bench 3.0 Long-horizon test, K3.1 Pro scored 61.2%, up from K3's 55.95%.
The picture is mixed: Moonshot's own comparison shows Kimi trailing Astra by 30 points and Opus by 25.2 points on a science test, and Moonshot and Artificial Analysis report different Terminal-Bench results.
The influencer claims add a credibility question. It is unclear whether the instructions came from Moonshot, an agency or another party, how many creators received them, or whether any paid posts ran, and no company response had been identified as of October 25. Such a campaign could raise disclosure issues under X's policy and Federal Trade Commission guidance, depending on the actual relationships involved. Artificial Analysis says the allegations do not by themselves undermine its benchmark results. Moonshot attributes the upgrade to substantially more computing in post-training and extensive reinforcement learning, in which a model improves from feedback on attempted tasks, but has not described a new architecture. The API models are `kimi-k3.1-pro` and `kimi-k3.1-pro-highspeed`, and enterprise administrators must switch on access because it starts disabled. Self-hosting is not yet possible while final weights and license terms are pending. Organizations with data-location or government-contract requirements should verify where their data is processed; a Department of Homeland Security order excludes DeepSeek from civilian federal procurement but does not target Kimi, though it reflects tighter scrutiny of Chinese-developed AI. For businesses, a bounded trial on their hardest recurring tasks, tracking accepted results, total cost and time saved, is the practical next step.
