Kimi Team: K2 MuonClip training and agentic post-training

What changed

Combined QK-clipped Muon optimization with large-scale expert-model training and synthetic/interactive agentic post-training.

Scientific scope

Combined QK-clipped Muon optimization with large-scale expert-model training and synthetic/interactive agentic post-training. Muon, MLA and sparse expert components are inherited. MuonClip and the integrated K2 training recipe are scoped here; benchmark claims are not independent validation or every K2 variant.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Muon, MLA and sparse expert components are inherited. MuonClip and the integrated K2 training recipe are scoped here; benchmark claims are not independent validation or every K2 variant.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Kimi K2: Open Agentic Intelligence
• Kimi Team: K2 MuonClip training and agentic post-training — official team research record

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction