GLM-5.2
Z.ai's flagship GLM with, for the first time, a solid 1M-token context; 62.1 on SWE-Bench Pro and 40.5 on HLE, MIT-licensed, weights following three days after subscriber launch.
A 753B-parameter MoE (per the HF card), with IndexShare cutting per-token FLOPs 2.9x at 1M context. Z.ai says it trails Claude Opus 4.8 by about 1% on FrontierSWE. Went to Coding Plan subscribers on 2026-06-13; API and weights on 2026-06-16.
- Date
- Saturday, 13 June 2026
- Lab
- Zhipu AI / Z.ai
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| SWE-bench Pro | 62.1 vs GLM-5.1 58.4 | company |
| Terminal Bench 2.1 | 82.7 best reported harness; 81.0 with the Terminus-2 harness vs Opus 4.8 85.0 (HF card and docs agree) | company |
| GPQA-Diamond | 91.2 | company |
| Humanity's Last Exam | 40.5 | company |
| Context window | 1M tokens | company |
2026-06-13 subscriber launch per aggregator reports, 2026-06-16 per Z.ai release notes and HF repo. Terminal Bench 2.1 differs between HF card (82.7) and docs (81.0). A claimed Artificial Analysis Intelligence Index of 51 comes only from secondary reports.
Sources
- huggingface.co/zai-org/GLM-5.2
- docs.z.ai/guides/llm/glm-5.2
- docs.z.ai/release-notes/new-released
- docs.apiyi.com/en/news/glm-5-2-launch
- simonwillison.net/2026/Jun/17/glm-52/
This record was checked and corrected against its sources on 6 October 2026. How we check