Head to head · Work · updated Oct 27, 2026
Docling vs Surya
Docling scores 64 to Surya's 42 on public evidence for extract data from documents, and they rank #2 and #3 there. Both are free to start. Numbers below are the same public signals the ranking uses, recomputed nightly — nothing here is sponsored.
Docling
Turns PDFs, Office files and scans into structured markdown or JSON, layout and tables intact.
+ Reads a table as a table, and does it on your own machine with nothing sent anywhere.
− Heavy scanned documents want a GPU; CPU-only runs are slow on long PDFs.
Surya
OCR, layout and reading-order models covering 90+ languages, weights included.
+ Beats the old OCR engines on messy scans and non-Latin scripts, locally.
− The licence is not plain open source above a revenue line - read it before shipping.
Side by side
public evidence| Docling | Surya | |
|---|---|---|
| Evidence score / 100 | 64 | 42 |
| GitHub stars | 69,684 | 21,542 |
| Hacker News stories, 30d | 1 | 0 |
| Release-feed mentions, 30d | 0 | 0 |
| Last activity | 0 days ago | 4 days ago |
| Pricing | Open source · MIT; runs locally, no API key | Open source · Weights are free for research and small companies; a commercial licence applies above a revenue threshold |
| Platforms | cli, api, linux, mac, windows | cli, api, linux, mac |
| Maker | docling-project (IBM Research) | Datalab |
Which one?
Pick Docling if
- you want the stronger public evidence (64 vs 42)
- you need it on windows
- a bigger GitHub community matters (69,684 vs 21,542 stars)
Best for feeding your own documents to a model without uploading them.
Pick Surya if
Beats the old OCR engines on messy scans and non-Latin scripts, locally.
Best for scanned archives in languages Tesseract mangles.