---
title: "DeepSeek V4 Flash 0731 Released: 304B Parameters with Enhanced Agentic Capabilities and Strong Value"
source: "Simon Willison 博客（Atom）"
source_tier: "t2"
url: "https://simonwillison.net/2026/Jul/31/deepseek-v4-flash-0731/#atom-everything"
permalink: "https://feed.astralor.com/items/itm_2a97a806a43b15611e83f30b"
published: "2026-07-31T23:59:44.000Z"
score: 77
channel: "ai"
tags: ["Agentic Capabilities","DeepSeek","DeepSeek-V4-Flash-0731","OpenRouter","Large Language Models"]
---

# DeepSeek V4 Flash 0731 Released: 304B Parameters with Enhanced Agentic Capabilities and Strong Value

## AI commentary

> The real story here isn't parameter count—it's that a 304B model can beat larger rivals while being this cheap. For developers, this signals a value inflection point: building agentic apps may no longer require chasing ever-larger models. Simon's note about reasoning effort is a useful reminder that benchmark results don't automatically translate to great outputs; tuning matters before you judge a model.

## At a glance

DeepSeek has released the latest version in its V4 series, DeepSeek-V4-Flash-0731, with 304 billion parameters and a 167GB Hugging Face checkpoint. The company says it comes with substantially enhanced agentic capabilities. Artificial Analysis ranks it ahead of the 428B-parameter MiniMax M3 on the Intelligence Index. With pricing at $0.14 per million input tokens and $0.27 per million output tokens, it currently looks like one of the best value-per-intelligence models available. Simon Willison's tests found the default reasoning level produced disappointing results for a pelican prompt, but bumping reasoning effort to high via OpenRouter yielded a much better output. The model appears to punch well above its weight, though real-world performance depends heavily on reasoning effort settings.

## Key points

- DeepSeek-V4-Flash-0731 is a 304B-parameter model with a 167GB Hugging Face checkpoint, and DeepSeek highlights substantially enhanced agentic capabilities.
- Artificial Analysis ranks it ahead of the 428B-parameter MiniMax M3, suggesting it punches well above its weight.
- API pricing of $0.14 per million input tokens and $0.27 per million output tokens makes it a strong value-for-intelligence option.
- Simon Willison found default reasoning effort disappointing; raising the effort level produced significantly better results.

## Full text

deepseek-ai/DeepSeek-V4-Flash-0731 The latest release in DeepSeek's V4 family, "with substantially enhanced agentic capabilities". It's 304 billion parameters - 167GB on Hugging Face - but it appears to punch well above its weight. Artificial Analysis rank it ahead of MiniMax M3 - a 428B model. It's $0.14/million input and $0.27/million output pricing means this may currently be the best value-per-intelligence model out there. It's looking very good on the Intelligence Index vs. Cost per Intelligence Index Task chart: I got a disappointing pelican from it using the default reasoning level via OpenRouter: But when I bumped reasoning level up to high I got something much better: llm -m openrouter/deepseek/deepseek-v4-flash-0731 -t pelican -o reasoning_effort high Via Hacker News Tags: ai, generative-ai, llms, pelican-riding-a-bicycle, deepseek, llm-release, openrouter, ai-in-china, artificial-analysis

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_2a97a806a43b15611e83f30b