Pyyan / Compare / Hy4 preview vs DeepSeek V4.1 Flash
Hy4 preview vs DeepSeek V4.1 Flash
2 of 5Open-Weight Models · verified 13 Sept 2026
Hy4 previewTencentcurrentDeepSeek V4.1 FlashDeepSeekcurrent
| Specification | Hy4 preview | DeepSeek V4.1 Flash |
|---|---|---|
| Summary | 770B parameters under Apache 2.0, and it helped optimise its own training. | Twice the parameters of the Flash it replaces, and cheaper to run. |
| Context | 1M | 1M |
| Input ($/Mtok) | $0.834 | $0.3 |
| Output ($/Mtok) | $2.501 | $1.2 |
| Licence | Apache 2.0 | MIT |
| Max output | 64K | 256K |
| Parameters | 770B total, 49B active | 552B total, 8B / 16B active |
| Released | 28 Aug 2026 | 10 Sep 2026 |
| Model ID | tencent/Hy4-preview | deepseek-flash |
| Category | Open-Weight Models | Open-Weight Models |
| Official | — | DeepSeek ↗ |
Highlighted rows are where these differ.
Hy4 preview
- 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.
DeepSeek V4.1 Flash
- 552B total, 8B active while reading the prompt and 16B while generating, with a 1M token context
- DeepSeek says it beats V4-Pro; V4-Pro requests route to it from 14 September 2026
- MIT licence, native image input, and a KV cache about a quarter the size of V4-Flash's
Best for cheap agentic coding with images.