← All models

DeepSeek OCR

DeepSeek · 2025 · Vision-language OCR

Token-efficient OCR for scans, photos, and mixed layouts. Run it locally in Delta when documents cannot leave the building.

Maker
DeepSeek
Architecture
Vision-language OCR
Fits from
Phone / edge · 6 GB+ RAM
Context
up to 131K
Runs
Local Intelligence
OCRVisionScansPhotos of textPrivate documents

Sizes

DeepSeek OCRQ6_K

requires 6 GB+ memory · 131K context · Phone / edge

2.6 GBWeights

Memory at context length

Approximate working set for DeepSeek OCR as context grows.

4K
3.2 GB
8K
3.4 GB
16K
3.7 GB
33K
3.9 GB
66K
4.1 GB
131K
4.3 GB