DeepSeek Unveils V4 Flash and V4 Pro a Year After Its First Wave

DeepSeek released previews of V4 Flash and V4 Pro on April 24, roughly a year after the first DeepSeek release reset US tech stock prices for a day. Bloomberg’s coverage called the timing pointed. This one shipped quietly into a market that had already absorbed the lesson.

Flash and Pro split the workload the way other model families now do. Flash handles speed and cost. Pro handles harder coding and reasoning, with what DeepSeek claims are real gains on agentic benchmarks. The company didn’t release detailed scorecards alongside the preview, which is normal for a preview release.

For AI visibility purposes the picture is mostly background context. DeepSeek’s models are widely used inside China and across parts of APAC, but in US-facing AI search surfaces, ChatGPT, Gemini, Perplexity, and Copilot still dominate citation volume. The exception is businesses with multilingual audiences or APAC operations, where DeepSeek’s presence in localized AI tools is a citation channel worth knowing about.

The bigger signal is that the model release calendar keeps compressing. GPT-5.5 on April 23. DeepSeek V4 on April 24. Two frontier-class previews inside 24 hours. Whatever stable picture you have of which model cites which sites is a snapshot, and the snapshots get older fast.

Sources: Bloomberg, April 24, 2026.