← Back to archive

Published fact-check

Kimi K2 Model Released by Moonshot AI

Claim checked

“Kimi K2.6 code preview is ou”

Published April 13, 2026 at 8:48 PM

Verdict

Mixed

Moonshot AI has released Kimi K2, a large-scale Mixture-of-Experts (MoE) language model with 1 trillion total parameters. While the user claim mentions a "Kimi K2.6 code preview," official documentation and repository data currently identify the latest major release as Kimi K2, which includes specialized variants for base research and instruction-based agentic tasks.

1 reviewed source behind this verdict.

Reasoning

The evidence confirms the release of the Kimi K2 series by Moonshot AI. The model features 32 billion activated parameters and is optimized for coding and agentic intelligence, achieving high scores on benchmarks like SWE-bench and LiveCodeBench. However, the specific version number "K2.6" mentioned in the social media post is not supported by the official GitHub repository or technical documentation, which refers to the model simply as Kimi K2. It is possible the user is referring to a specific iteration or preview not yet reflected in the primary documentation, but based on the available evidence, only Kimi K2 is verified.

Source quality: The official GitHub repository for Moonshot AI's Kimi K2 provides comprehensive technical details, benchmarks, and deployment guides, though it does not mention a 'K2.6' version specifically.

Key checks

  • Existence of Kimi K2: Moonshot AI has officially released Kimi K2, a 1-trillion parameter MoE model with strong coding capabilities.

  • Version Number Discrepancy: Official documentation identifies the model as Kimi K2 (Base and Instruct variants); there is no mention of a 'K2.6' version in the technical report or repository.

  • Coding Performance: Kimi K2 shows state-of-the-art performance in coding tasks, scoring 65.8% on SWE-bench Verified (Agentic Coding).

Confidence

High