GLM-4.6
Coding-focused refresh of GLM-4.5 with 200K context, claimed on par with Claude Sonnet 4 and over 30% more token-efficient.
357B-parameter MIT-licensed MoE with context raised from 128K to 200K and tool use during reasoning. Zhipu published test prompts and agent trajectories for its Claude Code coding comparisons so others could check the claims.
- Date
- Tuesday, 30 September 2025
- Lab
- Zhipu AI / Z.ai
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Token efficiency vs GLM-4.5 | over 30% better company-reported | company |
| Context window | 200K (from 128K) | company |
Release date from Z.ai release notes (HF repo created 2025-09-29 UTC).
Sources
This record was checked against its sources on 6 October 2026. How we check