deepseek-ai/DeepSeek-V4-Flash-0731

Simon Willison · Simon Willison · 2026-07-31

DeepSeek releases V4-Flash-0731, a 304-billion-parameter model with enhanced agentic capabilities that Artificial Analysis ranks ahead of a larger 428B rival, at a price point Willison describes as potentially the best value-per-intelligence ratio currently available.

Open original ↗

Appears in

Kimi K3 and Qwen3.8-Max: Chinese Labs Close Gap with Closed Frontier Models

Extraction

Topics: llm-releasemodel-pricingopen-weight-modelsdeepseekmodel-benchmarks

Claims

DeepSeek V4-Flash-0731 has 304 billion parameters and is positioned as having substantially enhanced agentic capabilities.
Artificial Analysis ranks V4-Flash-0731 ahead of MiniMax M3, a 428B model, on its Intelligence Index.
At $0.14/million input and $0.27/million output tokens, it may offer the best value-per-intelligence ratio of any model currently available.
The model produces significantly better outputs when reasoning level is set to high rather than left at default.

Key quotes

with substantially enhanced agentic capabilities

this may currently be the best value-per-intelligence model out there