DeepSeek-OCR (LT) — OCR AI Model
DeepSeek OCR reads handwriting, mixed-language pages and awkward layouts that trip lighter engines, returning markdown beside a text-overlay image. GPU-heavy, thirty to sixty seconds; worth it on a hard page.
What it does
DeepSeek OCR reads handwriting, mixed-language pages and awkward layouts - the pages that trip the lighter engines - and returns markdown alongside an image with the text located on it.
It is the heavy option: GPU-bound, thirty to sixty seconds, and worth that on a hard page rather than a clean one. For clean printed Korean, PaddleOCR is faster; for tables and formulas, GLM-OCR.
When to use it
- Handwriting, which the lighter OCR engines will not read
- A page mixing two languages, detected rather than declared
- Tables, figures and mixed-format pages inside one document
- Markdown out, with headings and paragraphs still standing
- Finding where a particular string sits, by bounding box
Details
- CategoryOCR
- Year2025
- LicenseMIT
Compliance & Provenance
- ProviderOpen-source · Specialized
- EU AI Act RiskMinimal Risk
- Art. 50 TransparencyNot applicable
Inputs & Outputs
- ImageInputimage
Source image containing text to recognize
- ImageOutputimage
Detected texts as an image
- TextOutputstring
Detected texts as markdown
Tags
Alternatives in Text Recognition(OCR)
- GLM-OCR
Document OCR with layout/table/formula. 10-40s.
- PaddleOCR
Printed-text OCR for Korean, Japanese, Chinese and English. Fast.
- Sarashina2.2-OCR
Japanese/English document OCR, vertical text.