DeepSeek V4.1 Flash hits 1,286 tokens per second in tests
DeepSeek V4.1 Flash demonstrated a text generation speed of 1,286 tokens per second at low per-task cost. In tests it produced a 3D mechanical watch diagram, a Rubik's Cube simulation, the Bean Blitz game and the GridLine and Folio apps.
- Text generation speed reaches 1,286 tokens per second
- Model built 3D watch, Rubik's Cube and Fall Guys clone
- Supports local deployment on consumer-grade GPUs
- Claimed low per-task cost for developers
Read next
AI