Hy4 flash Base vs Olala Mini 7A1B
Compare documented capabilities and access. A dash means a comparable figure is unavailable.
| Specification | ||
|---|---|---|
| Task | Multimodal pretraining | Multilingual language generation |
| Published capabilities | Pretrained checkpoint for continued training and adaptation. The Flash architecture uses a sparse Engram memory module. | Compact mixture-of-experts model combining hybrid attention and Mamba layers. |
| Model size | Approximately 220B total / 12B active per token | 7B total / 1B active |
| Access | Downloadable under the Tencent Hy License | Downloadable checkpoint for local inference or fine-tuning |
| Sources | Model weights | Model weights · Additional documentation |
| Inputs | Text, Images | Text |
| Context window | — | 32,000 tokens |
| Weights | Open weights · Tencent Hy License | Open weights · Apache-2.0 |
| Released | Nov 18, 2026 | Nov 19, 2026 |