Z.aiGLM-5 Flash
GLM-5.3-Flash
A native multimodal GLM model with sparse and linear attention.
Practical use
Practical guidance. Use screenshots and rendered output as evidence during visual development. Compare the requested behaviour with the visible result after each change. Treat the FlashX endpoint as a serving choice rather than a separate downloadable base.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API, Downloadable |
| Model identifiers | glm-5.3-flashglm-5.3-flashxglm-5.3-flashx-legalzai-org/GLM-5.3-Flashzai-org/GLM-5.3-Flash-BF16 |
| Architecture | 320 billion total parameters; 18 billion active. Uses manifold-constrained hyper-connections.GLM-5.3-Flash model card |
| Input and output | Video, images, text and files to text; one-million-token context and 128,000 output tokens.GLM-5.3 Flash and FlashX guide |
| Licence | MIT for the published weights.GLM-5.3-Flash model card |
| API variants | Flash and faster FlashX; the same guide also lists glm-5.3-flashx-legal.GLM-5.3 Flash and FlashX guide |