AITechNode5 months ago
DeepSeek releases OCR 2 with new visual encoding
1 min read
DeepSeek open-sourced DeepSeek-OCR 2, built on DeepEncoder V2 architecture. The model uses semantic reasoning instead of rigid scanning, needing 256 to 1,120 visual tokens per document. On OmniDocBench v1.5, it scored 91.09%, a 3.73% improvement over the previous generation.
Level
Hype check
Tap to vote and see what everyone thinks.
