GLM-4.5 and GLM-4.5-Air
355B-total (32B active) open MoE that unifies reasoning, coding and agent tool use in hybrid thinking and non-thinking modes, MIT licensed.
Pretrained on 23T tokens and post-trained with expert-model iteration and RL. A 106B/12B Air variant ships alongside. Zhipu says it ranks 3rd overall (2nd on agentic tasks) across 12 benchmarks, and sells the API at $0.2 input and $1.1 output per million tokens.
- Date
- Monday, 28 July 2025
- Lab
- Zhipu AI / Z.ai
- Kind
- open-weights
- Access
- open weights
- Price
- $0.2 input / $1.1 output per M tokens, 2025-07
Figures
| Measure | Value | Measured by |
|---|---|---|
| Overall rank across 12 benchmarks | 3rd (paper) / 2nd (docs) company-reported | company |
| SWE-bench Verified | 64.2 | company |
| TAU-Bench | 70.1 | company |
| AIME 24 | 91.0 | company |
Z.ai's docs say 2nd globally on the 12-benchmark average; the arXiv report (2025-08-08) says 3rd. Benchmarks are Zhipu-run. Release date from Z.ai's release notes.
Sources
- arxiv.org/abs/2508.06471
- huggingface.co/zai-org/GLM-4.5
- docs.z.ai/guides/llm/glm-4.5
- docs.z.ai/release-notes/new-released
This record was partly confirmed: some claims could not be checked on 6 October 2026. How we check