DeepSeek Ships V4-Flash, a 304B Agentic Model at Aggressive Pricing

DeepSeek released DeepSeek-V4-Flash-0731, the latest entry in its V4 family, described as having “substantially enhanced agentic capabilities.” The model is 304 billion parameters (167GB on Hugging Face).

Pricing is $0.14/million input and $0.27/million output tokens. Per Artificial Analysis benchmarks cited by the source, it ranks ahead of MiniMax M3, a larger 428B model, on intelligence-per-cost.