Overview
GLM-5 scales the GLM-4.5 design from 355B to 744B parameters (40B active) and increases pre-training data from 23T to 28.5T tokens. Zhipu trained it with slime, its asynchronous reinforcement-learning infrastructure, and targets agentic coding and long multi-step tasks.
Highlights
- 744B total, 40B active parameters
- About 200K-token context window
- Trained with large-scale asynchronous RL
- MIT license
- Available on the Z.ai API platform
Typical uses
Agentic coding, tool-using agents, research and long-document work.
Things to consider
- Self-hosting needs a multi-GPU cluster.
- Superseded by GLM-5.1 to 5.3 for many tasks.
Specifications are taken from the model card on Hugging Face and the provider’s announcement. Benchmark results depend on evaluation settings; see how AI benchmarks work.