Simon Willison • 7/31/2026

deepseek-ai/DeepSeek-V4-Flash-0731

This article reviews the release of DeepSeek-V4-Flash-0731, a 304 billion parameter model from DeepSeek's V4 family. It highlights the model's enhanced agentic capabilities, competitive performance against larger models like MiniMax M3 (428B), and its cost-effective pricing ($0.14/M input, $0.27/M output). The author notes strong results on the Intelligence Index and shares a personal test using OpenRouter, showing improved output with higher reasoning effort. This is relevant for AI developers and tech enthusiasts interested in cutting-edge models and cost-performance trade-offs.

0 comments

#llm #Reasoning #Pricing