DeepSeek: DeepSeek V4 Flash preview

by DeepSeek intelligent open-weight coding fast

DeepSeek V4 Flash is an open-weight MoE model with 284B total parameters (13B activated), featuring hybrid attention for 10x KV cache efficiency and a 1M token context window. Released in preview on April 24, 2026, it scores 7.0/10 (rank #97) and excels in high-throughput chat, coding, and agentic tasks at low cost.

Input: $0.10/1M Output: $0.20/1M Context: 1,048,576 tokens Rank: #130

Compare this model Creative showcases