DeepSeek Opens Time-Boxed V4.1 Flash Beta With Native Multimodal Architecture

DeepSeek has announced the launch of an internal beta for DeepSeek V4.1 Flash, an intermediate checkpoint based on a new model architecture with native multimodal support. This new release is framed as stronger, faster, and lower-cost compared to previous Flash versions, while maintaining Flash-tier pricing during the limited test window. Developers can access the preview without changing their existing API base URL. The new version natively handles multimodal input, allowing text and images to be processed within the same architecture, which reduces delays associated with converting images before they enter the language sequence.