The technical report for GLM-4.5, from Zhipu AI and Tsinghua University.
Architecture and training
An "open-source Mixture-of-Experts large language model with 355B total parameters and 32B activated parameters, featuring a hybrid reasoning method that supports both thinking and direct response modes."
Training: "multi-stage training on 23T tokens and comprehensive post-training with expert model iteration and reinforcement learning."
Reported results
The report organizes capability around ARC — agentic, reasoning, and coding:
| Benchmark | Score |
|---|---|
| TAU-Bench | 70.1% |
| AIME24 | 91.0% |
| SWE-bench Verified | 64.2% |
The efficiency claim is the report's emphasis: "with much fewer parameters than several competitors, GLM-4.5 ranks 3rd overall among all evaluated models and 2nd on agentic benchmarks." The agentic placement above the aggregate is the notable asymmetry — this is a model whose relative strength is in tool use and multi-step tasks rather than raw capability.
Two models were released: GLM-4.5 at 355B and GLM-4.5-Air at 106B.
Relationships
- supports: GLM-4 generation (GLM-4 through GLM-4.7) — the technical documentation for the release
- related: GLM-5: from Vibe Coding to Agentic Engineering (Zhipu AI, 2026) — the successor report
- related: Open-Weight Frontier Models, Mixture of Experts (MoE), Zhipu AI, US-China AI Competition: Different Races, Different Metrics