DeepSeek Launches DeepSeek-V4-Flash-0731 with High Efficiency and Agentic Capabilities
DeepSeek has released DeepSeek-V4-Flash-0731, a 304-billion-parameter model that offers enhanced agentic capabilities. Despite its massive size, it is priced extremely low at $0.14 per million input tokens and $0.27 per million output tokens. This release sets a new benchmark for the intelligence-to-cost ratio in the LLM market, outperforming larger models like MiniMax M3 at a fraction of the cost. It makes highly capable, agent-oriented AI applications significantly more accessible and cost-effective for developers. The model weighs 167GB on Hugging Face and features adjustable reasoning levels, where setting the reasoning effort to high significantly improves output quality for complex tasks. According to Artificial Analysis, it sits alone in the most attractive quadrant of the intelligence-versus-cost index.
## BACKGROUND
"Agentic capabilities" refer to an LLM's ability to act autonomously, use tools, and execute multi-step workflows to achieve specific goals. Artificial Analysis is an independent platform that benchmarks large language models based on speed, cost, and various intelligence metrics.