Z.aiGLM-5
GLM-5
A sparse text model for reasoning and software engineering.
Practical use
Practical guidance. Capacity planning must use both the total weight size and the active computation. For an engineering agent, also measure time spent in external tools, which model-only speed figures do not include.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | glm-5zai-org/GLM-5 |
| Architecture | 744 billion total parameters; 40 billion active. Integrates DeepSeek Sparse Attention.GLM-5 model card |
| Training | The card reports 28.5 trillion pre-training tokens.GLM-5 model card |
| Licence | MIT.GLM-5 model card |
| Access | Current API listing and published weights.Current Z.ai API catalogue |
Related models
A text model with stronger post-training on the GLM-5.2 base.
Launch date not confirmedAPI · DownloadableA long-context GLM release with a shared sparse-attention indexer.
Launch date not confirmedAPI · DownloadableA GLM release trained for sustained multi-step work.
Launch date not confirmedAPI · Downloadable