DeepSeek Teases Vision for V4

This looks less like a launch announcement than a roadmap leak, but it matters because flagship vision support is now table stakes for top-tier AI labs.

• –DeepSeek’s official V4 preview, published April 24, 2026, is still text-only and says multimodal capabilities are still being worked on.
• –DeepSeek already has a vision-language track record with VL and OCR models, so adding vision to V4 would consolidate capabilities into the flagship line.
• –Community reaction is skeptical of separate vision side models; users want native multimodality, not a bolted-on extra model.
• –If vision lands first in V4-Lite or V4.1, that would fit the pattern implied by community discussion and leaked interface hints.
• –For developers, native vision in the main model would make document parsing, OCR, image understanding, and multimodal agents simpler to ship.