DeepSeek V3 is a 685B-parameter Mixture-of-Experts model and the newest generation in DeepSeek’s flagship chat model family. As the successor to the earlier DeepSeek V3, it delivers strong performance across a wide range of tasks.
| Creator | Deepseek |
| Release Date | March, 2025 |
| License | MIT |
| Context Window | 128,000 |
| Image Input Support | No |
| Open Source (Weights) | Yes |
| Parameters | 671B, 37B active at inference time |
| Model Weights | Click here |
