DeepSeek released DeepSeek-V4.1-Flash, a 552B-parameter Mixture-of-Experts open model with a one-million-token context window, multimodal text-image input, and a dramatically reduced global KV cache of 890 bytes per token. The model uses a Causal Encoder-Decoder design with only 8B parameters active during prefill and 16B during decoding, plus Compressed Sparse Attention 2, FP4 KV caching, and hierarchical indexing to cut compute and memory needs for long prompts. Additional features such as Engram conditional memory, SWA Bounded Replay, Single-Pass mHC, and DSpark speculative decoding aim to lower infrastructure costs, speed inference, and support long-running AI agents that process very large contexts efficiently.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.