DeepSeek releases V4 Pro and V4 Flash in preview
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- DeepSeek rolled out preview versions of its V4 Pro and V4 Flash flagship models on April 24, 2026, with the company's own framing that V4 Pro trails the performance of state-of-the-art models by about 3 to 6 months
- DeepSeek published a companion paper titled "DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence" on Hugging Face, positioning the model's differentiator around million-token context efficiency rather than raw frontier performance
- DeepSeek released the models as open-source, a framing the New York Times, The Next Web, and CNBC each emphasized in their coverage of the launch
- Barron's Online recast the launch as a competitive blow to Nvidia, arguing that DeepSeek's release lays bare a missed opportunity for Nvidia's AI chips in a key market
- South China Morning Post reported the launch triggered an investor rotation into Chinese chipmakers, reflecting a market response to DeepSeek's renewed presence in the AI race
- Nikkei Asia, Reuters, France 24, and Al Jazeera each framed the release as an escalation in US-China AI competition, describing the V4 launch as a fresh challenge to US rivals roughly a year after DeepSeek's prior viral rise
Why it matters: DeepSeek's own admission that V4 Pro trails state-of-the-art by 3 to 6 months tempers the "China catches up" narrative that drove last year's stock shock, but the open-source release and the million-token-efficiency emphasis still give Chinese chipmakers and DeepSeek's developer community a new entry point — a point Barron's and the South China Morning Post both ran with, even as most US-China-competition coverage glossed over the performance gap.
