Pyyan / Compare / GLM-5.3 vs Atria Dawn Preview
GLM-5.3 vs Atria Dawn Preview
2 of 5Open-Weight Models · verified 16 Sept 2026
GLM-5.3Zhipu AIcurrentSAtria Dawn PreviewShanghai AI Laboratorycurrent
| Specification | GLM-5.3 | Atria Dawn Preview |
|---|---|---|
| Summary | Post-trained from the GLM-5.2 base, with the weights following two weeks later. | 744B agentic weights under the MIT licence, published without a launch. |
| Context | 1M | 256K |
| Input ($/Mtok) | $1.4 | Not published |
| Output ($/Mtok) | $4.4 | Not published |
| Licence | Open weights | MIT |
| Max output | 32K | Not published |
| Parameters | Not published | 744B total |
| Released | 14 Aug 2026, weights 27 Aug | 14 Sep 2026 |
| Model ID | zai-org/GLM-5.3 | internlm/Atria-Dawn-Preview |
| Category | Open-Weight Models | Open-Weight Models |
| Official | — | Shanghai AI Laboratory ↗ |
Highlighted rows are where these differ.
GLM-5.3
- Post-trained from the GLM-5.2 base rather than trained from scratch
- $1.40 in and $4.40 out per million tokens, with a 1M token window
- Announced 14 August, weights published 27 August
Best for the full-size open flagship.
Atria Dawn Preview
- Built on Zhipu's open GLM-5.2 rather than trained from scratch
- 96.0 on DeepSearchQA and 86.5 on CyberGym, both reported by the lab
- BF16 and FP8 checkpoints, served through SGLang or vLLM
Best for agentic work on your own cluster.