DeepSeek V4-Flash: Agent Leap, Open Source, Ultra-Low Price

On July 31, DeepSeek officially released the V4-Flash model API and open-sourced the weights on Hugging Face. The update does not change the architecture but significantly improves agent capabilities through re-post-training, with benchmark scores surpassing the previous V4-Pro-Preview, and offers flagship-level performance at a highly competitive price, sparking community discussion.

Confirmed

• API and Architecture: The DeepSeek-V4-Flash official API is now in public beta. The model uses a MoE architecture, supports 1M token context, natively supports the OpenAI Responses API format, fully adapts to Codex, and includes a one-click configuration script. The official clarification states that this upgrade is limited to the V4-Flash API; V4-Pro API and App/Web remain unchanged, with V4-Pro official release coming soon.
• Agent Capability Leap: Officially confirmed, agent capabilities have greatly improved: TerminalBench2.1 score 82.7 (up 25.8 points from the April preview), Cybergym 76.7, Toolathlon 70.3, DeepSWE 54%. It scores 50 on the Artificial Analysis Intelligence Index, up 10 points from the previous generation, surpassing its own V4-Pro by 6 points, and approaching GPT-5.6 Luna and GLM-5.2.
• Open Source and Ecosystem: DeepSeek released DeepSeek-V4-Flash-0731 weights on Hugging Face under the MIT license, supporting 8-bit and fp8 precision. Hugging Face engineer Victor Mustar deployed a free public inference endpoint that requires no token. Coding tool Cline has already integrated the model.
• Extreme Cost-Effectiveness: The pricing is highly competitive: $0.11 per million input tokens and $0.27 per million output tokens. According to netizen comparisons, the cost is only 1/50th of Kimi K3 or 1/30th of Gemini 3.6 Flash output price.

Unconfirmed

• The DeepSWE benchmark result is currently claimed solely by DeepSeek; the poster emphasizes it has not been fully independently verified by third parties.

Why it matters

• V4-Flash achieves a leap in agent capabilities through re-post-training, providing developers with a more powerful model for task execution and tool calling. Its native Codex compatibility, open-source weights, and highly competitive cost further lower the barrier for agent development, seen by some observers as another "DeepSeek moment."

2026-07-31 ~ 2026-08-01 · 100 related posts

Full story(6 episodes)→

• Episode 1: DeepSeek V4 Flash Hands-on: Low Cost, High Performance Sparks Community Buzz(2026-07-31, 33 posts)
• Episode 2: DeepSeek V4-Flash: Agent Leap, Open Source, Ultra-Low Price(2026-07-31, 100 posts)
• Episode 3: DeepSeek V4 Series Sparks Debate: Flash Performance Gains and Pro Scale(2026-07-31, 3 posts)
• Episode 4: DeepSeek V4 Flash Tops OpenRouter, Costs 1% of Rivals(2026-08-03, 14 posts)
• Episode 5: DeepSeek-V4-Flash Released Across Major Platforms(2026-08-06, 4 posts)
• Episode 6: DeepSeek-V4 Flash Benchmarked: One-Sixth Cost, 80% of Luna's Performance(2026-08-07, 14 posts)

• DeepSeek-V4-Flash Officially Released, Surpassing Pro Preview in Agent Benchmarks — 赛博禅心 · 2026-07-31
• [source] DeepSeek updates V4-Flash, teases imminent V4-Pro release — Nunki08 · 2026-07-31
• DeepSeek-V4-Flash Updated with Significant Score Improvements — cedric_chee · 2026-07-31
• DeepSeek V4 Flash Model Goes Live on Official API — ElMess-Siah · 2026-07-31
• Updated DeepSeek V4 Flash Scores 54% on DeepSWE Benchmark — zainhas · 2026-07-31
• DeepSeek Clarifies V4-Flash Upgrade Details, V4-Pro Coming Soon — deepseek_ai · 2026-07-31
• DeepSeek V4-Flash Natively Supports Codex with Setup Guide — zephyr_z9 · 2026-07-31
• DeepSeek Quietly Launches V4-Flash API; Open Weights Expected — Hot_Example_4456 · 2026-07-31
• DeepSeek-V4-Flash Tested: Blazing Fast Speed and Outperforms Pro — vista8 · 2026-07-31
• [source] DeepSeek Releases New DeepSeek-V4-Flash-0731 Model — deepseek-ai · 2026-07-31
• [source] DeepSeek V4-Flash Scores 50 on Artificial Analysis Index, 1 Point Below GLM-5.2 — MagicZhang · 2026-07-31
• DeepSeek V4-Flash Hits Index Score of 50 at a Cost of Just $0.20 — xeophon · 2026-07-31
• DeepSeek-V4-Flash Benchmarks Leak: Higher IQ Density but Questionable Pricing — teortaxesTex · 2026-07-31
• DeepSeek's new model shows abnormal cache write costs, beating GPT without them — teortaxesTex · 2026-07-31
• DeepSeek cost anomaly: 50x concurrent pricing could break all curves — xeophon · 2026-07-31
• DeepSeek-V4 Flash Scores 50 on AA Index, Beating Price-Cut GPT 5.6 in Cost-Efficiency — zainhas · 2026-07-31
• DeepSeek's Price War Sparks Hype for DeepSeek 4 Pro — kimmonismus · 2026-07-31
• DeepSeek-V4-Flash-0731 far surpasses DeepSeek-V4-Pro-Preview in benchmarks — SnooBunnies8392 · 2026-07-31
• DeepSeek-V4-Flash surpasses V4-Pro-Preview in latest benchmarks — Outside-Risk-8912 · 2026-07-31
• Rumor: DeepSeek-V4 Performance to Rival Claude Opus 4.8 with Disruptive Pricing — teortaxesTex · 2026-07-31
• DeepSeek-V4-Flash API Launches in Public Beta with Upgraded Agent Capabilities — gaganghotra_ · 2026-07-31
• DeepSeek V4 Flash Agent Capabilities Surge at 1/100th of Competitors' Cost — teortaxesTex · 2026-07-31
• DeepSeek V4 Flash 0731 Scores 50 on Intelligence Index, Closing Gap with GPT-5.6 — ArtificialAnlys · 2026-07-31
• DeepSeek Models 17-100x Cheaper Than Moonshot, Dominating Long-Context Costs — teortaxesTex · 2026-07-31
• DeepSeek v4 Update Closes Gap with GLM 5.2 in Just 1.5 Months — cedric_chee · 2026-07-31
• DeepSeek v4 Flash Scores 50 on AA Index, Closing Gap with Opus at Unbeatable Price — kimmonismus · 2026-07-31
• DeepSeek-V4-Flash Counters OpenAI's Price War with Extreme Cost-Effectiveness — teortaxesTex · 2026-07-31
• DeepSeek V4 Flash API Public Beta: Scores Near Opus 4.8, Natively Integrates with Codex — AI寒武纪 · 2026-07-31
• DeepSeek V4 Flash Hits the Pareto Frontier for Cost vs. Intelligence — PMinervini · 2026-07-31
• DeepSeek-V4-Flash Agent Performance Leaps, Rivals Opus and GLM — mark_k · 2026-07-31
• DeepSeek-V4-Flash Agent Benchmarks Surge, Matching Frontier Models — kimmonismus · 2026-07-31
• DeepSeek-V4-Flash Intelligence Index Jumps, Unbeatable Cost Per Task — eyishazyer · 2026-07-31
• DeepSeek-V4-Flash Released: Flagship Performance at 1/100th the Cost — TheZachMueller · 2026-07-31
• DeepSeek-V4-Flash Surfaces on Hugging Face with Million-Token Context — victormustar · 2026-07-31

16 near-duplicate retellings: APPSO · 歸藏的AI工具箱 · op7418 · zainhas · 数字生命卡兹克 · jiqizhixin · 卡尔的AI沃茨 · 智东西 · ricklamers · 新智元 · 机器之心 · haider1 · Boring_Aioli7916 · cedric_chee · FinanceYF5 · victormustar