DeepSeek-V4

DeepSeek-V4 is an open-source series of Mixture-of-Experts (MoE) language models designed for ultra-long context processing, supporting up to 1 million tokens by default. It includes two variants: V4-Pro with 1.6 trillion parameters and V4-Flash with 284 billion parameters, both leveraging a novel hybrid attention architecture to reduce computational and memory costs. This collection represents a significant step in democratizing large-scale AI models for developers and researchers.