DeepSeek-ai released DeepSeek-V4-Flash-0731, a 304 billion parameter model that costs $0.14 per million input tokens and $0.27 per million output tokens. This pricing structure places it ahead of the MiniMax M3 model on the Artificial Analysis Intelligence Index chart despite having fewer parameters. The release sits alone in the upper left quadrant of the cost versus intelligence graph, while competitors like Grok 4.5 and Gemini 3.6 Flash cost ten times more per task.
Testing via OpenRouter showed that default reasoning settings produced a poorly drawn pelican with floating bicycle parts. Increasing the reasoning effort to high generated a coherent image of the bird riding the bike correctly. The model demonstrates substantial agentic capabilities for its size and price point.
* 167GB model file size on Hugging Face
* $0.14 per million input tokens
* $0.27 per million output tokens
