B.AI Launches DeepSeek-V4.1-Flash Multimodal Model

B.AI has officially released DeepSeek-V4.1-Flash, a native multimodal MoE model featuring a 552 billion parameter backbone and a Causal Encoder-Decoder architecture. The model supports a 1 million token context window and 384K output, designed for long-cycle programming agents and high-concurrency systems. Access is now available via both Web Chat and API.