On October 23, 2026, Kimi's official account posted an about 24-second teaser video. Reports from IT Home, Phoenix Tech, and other Chinese outlets describe an old-style computer running Kimi Code, with the model field showing "Model: Kimi K3.1." Moonshot AI has not published a formal launch date, model size, or price list with that clip.

Key Facts

Details
What appeared Official Kimi teaser video, about 24 seconds
Date reported October 23, 2026
On-screen label Model: Kimi K3.1 inside Kimi Code (per IT Home and matching reports)
Formal release Not announced
Specs / pricing Not published
Public API Not confirmed as generally available
Prior model Kimi K3 launched in July 2026 with a 1M-token context window

Background

Kimi K3 shipped in July 2026 with a stated 1 million token context. Later product integrations that are on the public record include GitHub Copilot adding Kimi K3 on August 6, 2026, and Amazon's Kiro coding assistant announcing K3 support on October 15, 2026, with reporting that Kimi was the first open model in a global revenue-share setup on AWS Bedrock.

Community and code trails also pointed at a K3.1 label before the teaser. Developers reported a kimi-k3-1 entry in Moonshot API back-office registries, and on October 19 the SGLang repository gained a commit titled along the lines of "Kimi-K3.1 support for the K3 chat encoder." Those are community and codebase clues. They are not confirmed in an official Moonshot product brief for K3.1.

Reactions and Impact

Chinese tech media treated the clip as a pre-launch signal and noted that developer discussion followed quickly. No independent performance numbers for K3.1 were attached to the teaser itself.

What We Don't Know Yet

  • Official release date for Kimi K3.1.
  • Parameter count, architecture changes, and how it differs from K3.
  • Pricing, whether weights will be open, and whether a general public API will open on day one.
  • Whether features seen only in community code reviews (for example extra reasoning tiers) ship in the product.
  • Benchmark standing versus K3 and peer coding models.

Sources