DeepSeek-V4.1-Flash officially launches on B.AI, with the native multimodal MoE large model now offering dual access via Web Chat and API.

The B.AI platform announced that DeepSeek’s newly released native multimodal MoE large model, DeepSeek-V4.1-Flash, is now officially available. Featuring a 552 billion parameter backbone and a pioneering Causal Encoder-Decoder architecture, the model employs asymmetric computational scaling for inputs and outputs, significantly reducing operational costs while comprehensively outperforming the previous-generation flagship model.