GLM-5 Team: headwise optimization and asynchronous agent training

What changed

Adapted Muon updates by attention head, shared multi-token prediction parameters and decoupled rollout generation from training in an asynchronous agent-learning system.

Scientific scope

Adapted Muon updates by attention head, shared multi-token prediction parameters and decoupled rollout generation from training in an asynchronous agent-learning system. Inherits Muon, MLA, DSA and multi-token prediction; GLM5 does not invent those components. Original February report controls scope; later5.2/5.3 architecture and benchmark claims are not backdated. Asynchronous-learning and quality gains are source-reported.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits Muon, MLA, DSA and multi-token prediction; GLM5 does not invent those components. Original February report controls scope; later5.2/5.3 architecture and benchmark claims are not backdated. Asynchronous-learning and quality gains are source-reported.

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.

• GLM-5: from Vibe Coding to Agentic Engineering
• GLM-5 Team: headwise optimization and asynchronous agent training — official project

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