DeepSeek Releases V4 Flash Model with Enhanced Agentic Capabilities

DeepSeek has unveiled its latest model, DeepSeek-V4-Flash-0731, which is part of the V4 series and features significantly improved agentic capabilities. This model, containing 304 billion parameters and occupying 167GB on Hugging Face, is touted as a high-performance contender that exceeds expectations. As reported by Artificial Analysis, it outperforms the MiniMax M3, which has 428 billion parameters, in terms of intelligence. With competitive pricing at $0.14 per million input tokens and $0.27 per million output tokens, it stands out as potentially the best value for intelligence available. Simon Willison shared the news on 31st July 2026, noting that after testing via OpenRouter, the model's output improved dramatically when reasoning effort was set to high. Additionally, Willison offers a sponsorship option for $10 per month, which includes a monthly email digest of key LLM updates.