Kimi’s Chat, domestic Kimi Work, and Kimi Code now let you pick K3.1 Pro. In Code, the picker also lists K3.1 Pro Highspeed. OpenCode Go also lists Kimi K3.1 Pro as available (OpenCode announcement, about 02:29 Beijing time on October 24, 2026). Next to K3.1 Pro in Chat, the “进阶” label is a thinking-intensity setting, not a separate model name.
Kimi’s official X post and tech blog describe K3.1 Pro as a 3T-class model with 2.8T total / 104B active parameters (MoE), a 1M-token context window, and multimodal support. Product surfaces (Chat / Work / Code / API, plus Highspeed) are already live; open weights are planned for 2026-11-04 and are not released yet. Kimi’s posts claim it matches or approaches the strongest proprietary models on coding, agentic, and knowledge benchmarks; we have not re-tested those scores.
Chinese homepages name K3.1 Pro; English homepages have not updated yet. Related reading: Kimi K3.1 Has Shown Up in Kimi Code, but There's No Release Yet. For broader coding-agent context, see Best Coding Agents Right Now, Ranked by Artificial Analysis (October 2026).
Some help pages and platform model tables still list older lines such as K3, K2.8 Preview, and K2.7 Code HighSpeed. The Work landing page and help overview still lean on K3 wording. Kimi has published a standalone tech blog post for K3.1 Pro. Hugging Face still has no K3.1 weights entry (Moonshot’s latest public listing remains Kimi-K3). OpenRouter now lists `moonshotai/kimi-k3.1-pro` (about $2 / $10 per 1M tokens on the standard provider); see Kimi K3.1 Pro API Pricing (October 2026) for OpenRouter rates and the official platform table.
For running K3.1 Pro or K3.1 Pro Highspeed on real tasks, we have not tested them yet; if you have, you can send feedback through the site contact link in the footer.