DeepSeek V4 Flash API Launches in Public Beta With Major Agent Gains
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- DeepSeek rolled out the official V4 Flash API in public beta on July 31, 2026, touting enhanced agent capabilities and benchmark scores "far surpassing" the V4 Pro Preview version.
- DeepSeek clarified the V4 Flash-0731 update keeps the exact same model architecture and size as the preview; only the API was upgraded, while the V4 Pro API and App/Web models remain unchanged, with V4 Pro's official release "coming ASAP."
- Per analyst @haider1, the roughly 300B-parameter V4 Flash saw Terminal-Bench scores jump from 61.8 to 82.7 and DeepSwe from 7.3 to 54.4 — gains made at just $0.18 per million output tokens, roughly 100x cheaper than comparable frontier performance.
- BleepingComputer separately reported a hacker used DeepSeek AI to autonomously attack vulnerable servers, a security dimension running parallel to the benchmark hype.
- Superintelligence framed the release as DeepSeek "answering OpenAI's price cut overnight," positioning the timing as a direct competitive response rather than a standalone product launch.
- MarkTechPost and The Kaitchup highlighted the agentic and coding gains, with Kaitchup benchmarking it against smaller models to argue efficiency has overtaken raw scale in this cycle.
Why it matters: DeepSeek achieved major benchmark gains on V4 Flash through post-training alone at $0.18/m output tokens, undercutting Western frontier pricing by roughly two orders of magnitude. If independent testing confirms these scores, developers get near-frontier agentic coding capability for a fraction of current costs — while the parallel BleepingComputer report shows the same autonomous-agent strengths are already being weaponized against unpatched servers.
