DeepSeek Releases V4 Pro and V4 Flash Preview
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- DeepSeek released preview versions of V4 Pro and V4 Flash, with the company stating V4 Pro trails state-of-the-art models by about 3 to 6 months.
- DeepSeek published the V4 Pro paper on Hugging Face under the title "Towards Highly Efficient Million-Token Context Intelligence," highlighting the model's million-token context capability.
- Barron's reports Nvidia is "losing out" in a key China market as the V4 release exposes a missed opportunity for the chip leader.
- South China Morning Post reports investors are rotating toward Chinese chipmakers as DeepSeek intensifies AI competition, signaling domestic hardware momentum.
- The Independent reports DeepSeek claims V4 beats all open-source competitors, while Livemint claims the model surpasses ChatGPT and Gemini in key benchmarks.
- Multiple outlets — Reuters, The Information, Nikkei Asia and the NYT — frame the launch as a "long-awaited" return exactly one year after DeepSeek's viral "Sputnik moment."
- ZeroHedge, Nikkei Asia and France 24 cast the release as escalating the US-China compute and AI race, with China pushing open-source against US frontier labs.
Why it matters: DeepSeek's own framing — a 3-to-6-month gap to state-of-the-art — is more measured than the "beats ChatGPT" takes dominating secondary coverage, but it still marks a narrowing of the frontier gap from open-source China. The simultaneous Barron's/SCMP signal that Nvidia is ceding ground in China to domestic chipmakers means V4 lands as a dual shock: model parity pressure on US labs and chip-customer pressure on Nvidia.
