Pyyan / Open-Weight Models / Hy4 preview
Hy4 preview
770B parameters under Apache 2.0, and it helped optimise its own training.
currentTencentVerified 2026-09-02
770B
total, 49B active
Specifications
| Context | 1Msource · as of 2026-09-02 |
|---|---|
| Input ($/Mtok) | $0.834source · as of 2026-09-02 |
| Output ($/Mtok) | $2.501source · as of 2026-09-02 |
| Licence | Apache 2.0source · as of 2026-09-02 |
| Max output | 64Ksource · as of 2026-09-02 · low confidence |
| Parameters | 770B total, 49B activesource · as of 2026-09-02 |
| Released | 28 Aug 2026source · as of 2026-09-02 |
| Model ID | tencent/Hy4-previewsource · as of 2026-09-02 |
Official links
Hy4 preview model pageWhat matters
- 770B total parameters with 49B active, 78 layers, 256 routed experts plus one shared
- Apache 2.0, with an FP8 quantised build shipped alongside it
- Tencent used it during its own development, and report a 31.8% inference throughput gain from it
Best for the largest open weights.