Moonshot AI releases Kimi K3.1 teaser video, official launch date not yet announced
Moonshot AI has released a 24-second teaser video in which the model field in Kimi Code displays "Model: Kimi K3.1," but the official launch date, parameter scale, and pricing have not been announced. K3.1 had previously appeared in the API backend registry and the SGLang repository, supports a 1M-token context window, and adds a medium tier alongside the low, high, and max thinking-effort tiers. After its release, K3 was integrated into GitHub Copilot and AWS's Kiro, and the K3.1 teaser also centers on coding workflows.
Body · AI translation
Moonshot AI Releases Kimi K3.1 Teaser Video, Official Launch Date Not Yet Announced
2026-10-23 16:43 IT Home - Wen Zhou
Thanks to IT Home readers Qingchuan Wenhe, Agent, and Xiaoming Bu He Cha for the tips!
IT Home, October 23 — The official Kimi account today posted a 24-second teaser video, in which an old-fashioned computer is running Kimi Code, with the model field in the upper-left corner displaying "Model: Kimi K3.1." Combined with a series of earlier traces, all signs point to K3.1 being about to launch, and the developer community has already begun discussing it.
The earliest sign came in September this year, when multiple developers noticed that the identifier "kimi-k3-1" had appeared in the backend model registry of the Moonshot AI API platform, indicating that the model identifier had passed interface probing and could be called. On the same day, the Kimi official open platform also showed a preview of the K3.1 or K3 11111 model, supporting a 1M tokens context window.
In the early hours of October 19, a commit titled "Kimi-K3.1 support for the K3 chat encoder" appeared in the SGLang repository, adding K3.1 chat encoding adaptation to an experimental routing component, accompanied by 107 test cases.
According to a summary by Zhixingxing, K3 originally accepted three thinking-intensity levels: low, high, and max. K3.1 adds a medium level on top of that, while the routing default remains max; the K3.1 tokenizer file is byte-for-byte identical to K3's, reusing the same 16 special tokens, and the new tests cover chat formats, tool calls, and image input formats, constituting encoding-compatibility verification rather than performance benchmarking. An independent review by orcarouter further pointed out that the commit directly writes "moonshotai / Kimi-K3.1" and cites a specific model repository revision number, meaning that an identifiable K3.1 model entity is already in place on the Hugging Face side, just not yet public.
When Kimi K3 was released in July, it supported a 1 million Token context, and the official team said it would add low and high modes beyond the default max. If the SGLang code reflects final product behavior, K3.1's reasoning intensity will expand from three levels to four, inserting an intermediate level between low and high.
After K3's release, GitHub Copilot integrated Kimi K3 on August 6, and Amazon AWS's AI coding assistant Kiro announced integration with K3 on October 15, making Kimi the first open-source model to participate in global revenue sharing on AWS Bedrock. K3.1's choice of the Kimi Code interface as its teaser vehicle at least shows that coding and developer workflows remain one of the core narratives of this update. Beyond reasoning levels, K3.1's frontend clues also point to a 3D content creation entry called "make3d" and a multi-agent collaboration marker "OK_COMPUTER_SWARM," but these identifiers have not yet been officially confirmed by Moonshot AI.
As of IT Home's publication time, Moonshot AI has not yet announced K3.1's official release date, parameter scale, or pricing.