Model release · DeepSeek
DeepSeek-V4.1-Flash
- Model released
Vs V4-Flash: CED asymmetric architecture; native multimodal; KV cache about 1/4 HBM, 1/8 SSD; vendor claims multiple benchmarks beat V4-Pro; old deepseek-v4-flash / vision-exp temporarily routed here; deepseek-v4-pro remains a separate V4 Pro service (not permanently routed to Flash); off-peak about cache-hit ¥0.02 / miss ¥1 / output ¥4 (per million tokens), peak ×2
Availability
MIT open-source weights; API deepseek-flash; HF deepseek-ai/DeepSeek-V4.1-Flash; WorkBuddy/OpenCode