DeepSeek-V4: a million-token context that agents can actually use

Hugging Face · 5 months ago · 53

DeepSeek released V4 with two model variants—V4-Pro at 1.6 trillion total parameters (49 billion active) and V4-Flash at 284 billion total (13 billion active)—both supporting a 1 million token context window. At 1M tokens, V4-Pro requires 27% of the single-token inference compute of V3.2 and uses 10% of its KV cache memory, while V4-Flash drops to 10% of the FLOPs and 7% of the cache through hybrid attention mechanisms that compress key-value entries by 4x to 128x. The efficiency gains enable practical long-context agentic workflows where tool results accumulate across multiple rounds without exhausting GPU memory or degrading performance mid-task.