DeepSeek V4.1 Flash expands the open multimodal V4 line

DeepSeek V4.1 Flash coverage reports native multimodality, roughly 3x faster inference, a 70% API price cut, and benchmark gains, with users reportedly pushed from V4 Pro toward Flash. The earlier V4-Flash-Vision-Exp remains the more consequential public artifact, with approximately 305B MIT-licensed weights and vLLM/SGLang recipes; performance, lineage, and pricing claims require verification.