TODAY
DeepSeek launches image-understanding mode — hands-on feels like a new model
DeepSeek opened image-understanding mode to gradual rollout on April 30. Hands-on shows OCR, table recognition, and webpage-to-HTML are all usable. Looks like a model independent of V4 flash / pro — users found it knows April 2026 facts that V4 doesn't, though there's no official confirmation yet. Limits: a spatial reasoning task in thinking mode took 4+ minutes; colour-blindness tests still trip it up.
Published: 2026-05-01
Sources
- 量子位 — DeepSeek 識圖實測zh-CN
Tags
model-releasevisiondeepseekchina-ai