Here's the latest on the **DeepSeek Flash model** as of today, **July 31, 2026**: --- ## 🚀 DeepSeek-V4-Flash-0731 — Released Today! DeepSeek released **DeepSeek-V4-Flash-0731** on July 31, 2026, a major update to their V4-Flash model. Here are the key details: ### 📦 Availability - **Public beta** via the DeepSeek API — set model name to `deepseek-v4-flash` - **Same architecture** as V4-Flash-Preview (284B total params, 13B active, 1M context window) — only **re-post-trained** - **MIT-licensed weights** available on Hugging Face - Natively supports the **Responses API format** and is adapted for **Codex** - Legacy model names `deepseek-chat` and `deepseek-reasoner` were **discontinued on July 24, 2026** ### 🧠 Agent Benchmark Scores (huge improvements over V4-Pro-Preview) | Benchmark | Score | |---|---| | **Terminal Bench 2.1** | **82.7** | | **Cybergym** | **76.7** | | **Toolathlon (verified)** | **70.3** | | **DSBench-FullStack** | **68.7** | | **DSBench-Hard** | **59.6** | | **DeepSWE** | **54.4** | | **NL2Repo** | **54.2** | | **Agent Last Exam** | **25.2** | | **Automation Bench (Public)** | **25.1** | ### 📊 Competitive Positioning - **Artificial Analysis Intelligence Index score: 50** (up from 40 for the previous V4 Flash) - Nearly matches **OpenAI's GPT-5.6 Luna** (score of 51) — at roughly **60% lower cost** - Outperforms GPT-5.6 Luna on several agentic benchmarks ### 🔮 What's Next - The **official release of DeepSeek-V4-Pro** is expected to follow soon - The APP/WEB models remain unchanged for now — this update only affects the Flash API In short, DeepSeek-V4-Flash-0731 is a **highly competitive, cost-efficient model** that punches well above its weight class, especially in agentic/coding tasks.