A DeepSeek researcher just teased what comes next from the Chinese AI lab: a vision model.
Xiaokang Chen, who works at DeepSeek, posted a brief signal on April 28 suggesting a vision-capable model is in development. That's the full extent of what's been shared - no release date, no benchmark numbers, no architecture details. Just an early heads-up that it's coming.
DeepSeek's existing models - V3 and R1 - handle text and reasoning well enough to compete with ChatGPT and Claude on many tasks, while being available as open weights that anyone can run locally. Adding vision would mean the model could also understand images: read charts, interpret screenshots, describe photos, and process documents visually (rather than just extracting text from them).
That capability gap has been one of the main reasons developers still reach for GPT-4o when they need image understanding in their apps. An open-weight DeepSeek vision model would give the self-hosting crowd a serious alternative.
DeepSeek has a pattern of releasing models with little prior warning - a paper and weights drop simultaneously, with brief announcements. If that holds here, the gap between this tease and an actual release could be short. It could also be months. No timeline has been given, and a single researcher post is a long way from a product announcement.