TML-d24-varloop-research vs Hy4 Base
Compare documented capabilities and access. A dash means a comparable figure is unavailable.
| Specification | ||
|---|---|---|
| Task | Variable-depth reasoning research | Multimodal pretraining |
| Published capabilities | Researchers can vary the number of middle-layer passes with a single checkpoint. | Pretrained checkpoint for continued training and adaptation. Tencent’s flagship Hy4 uses a 78-layer language backbone with a separate vision encoder. |
| Model size | 26.4B total / 3.4B active parameters | 770B total / 49B active per token in the language backbone |
| Access | Downloadable research checkpoint | Downloadable under the Tencent Hy License |
| Operating limits | Native mixed-precision checkpoint occupies approximately 20 GB. | — |
| Sources | Model weights · Additional documentation | Model weights |
| Inputs | Text, Images | Text, Images, Video |
| Weights | Open weights · Apache-2.0 | Open weights · Tencent Hy License |
| Released | Nov 20, 2026 | Nov 20, 2026 |