DeepSeek released V4.1-Flash on September 10, introducing the smallest model in a new architecture family with native visual understanding. The company describes a 552-billion-parameter mixture-of-experts model built around a new Causal Encoder–Decoder architecture, with 8 billion active parameters for input processing and 16 billion for output.
The main engineering change is memory efficiency. DeepSeek says V4.1-Flash needs one quarter of the HBM and one eighth of the SSD storage for KV cache compared with the previous generation. Those are vendor-reported architecture figures; AiToolMap does not treat DeepSeek’s benchmark claims as independently reproduced performance evidence.
The model is live through the DeepSeek API as deepseek-flash with native multimodal support. DeepSeek has retired V4-Flash and V4-Flash-Vision-Exp, with their old API names temporarily routing to V4.1-Flash. More significantly, from 04:00 UTC on September 14, requests to deepseek-v4-pro will route to V4.1-Flash at V4.1-Flash rates until V4.1-Pro launches.
DeepSeek also changed API economics. Off-peak rates are half the peak rates, and the new pricing took effect at 04:00 UTC on September 10. The release therefore matters beyond a benchmark refresh: existing API integrations can receive a different underlying model through legacy names, while agent workloads gain a substantially smaller vendor-reported cache footprint.