Model components / Dataset Processing Tools

olmOCR

Converts PDFs and images to clean markdown/text using a 7B vision-language model, handling equations, tables, and handwriting. Emits Dolma-format documents; used in the Dolma 3 PDF-extraction path.

Ai2 olmOCR; repo active (v0.4.27, 2026-03-12), Apache-2.0 confirmed live 2026-07-21. High-profile document-extraction tool (19k+ GitHub stars); feeds the Dolma 3 pipeline.

Openness

5 high confidence

5.0

license: Apache-2.0(OSI)
source: public(allenai/olmocr)
governance: Allen Institute for AI
core-gated: ungated

Fully OSI-licensed (Apache-2.0), full source public.

https://github.com/allenai/olmocr/blob/main/LICENSE recorded 2026-07-21
Apache License Version 2.0 text

Adoption

3 high confidence

3.0

~22k PyPI downloads/month and 19k+ GitHub stars; a leading open document-extraction tool, used in the Dolma 3 data path.

https://pypistats.org/packages/olmocr recorded 2026-07-21
Downloads last month: 21,753

Capability

4 high confidence

4.0

Benchmark-anchored: leads open document-extraction tools on olmOCR-Bench but sits mid-pack on the neutral OmniDocBench, so near-SOTA rather than SOTA. Snapshot; extraction leaderboards move.

https://github.com/allenai/olmocr recorded 2026-07-21
olmOCR 2 (v0.4.0): 82.4 on olmOCR-Bench, ahead of Marker/MinerU/Mistral OCR/GPT-4o
https://github.com/opendatalab/OmniDocBench recorded 2026-07-21
OmniDocBench: olmOCR ~85.7, mid-pack behind MinerU2.5-Pro and Gemini-3 Pro

Unchanged since 2026-07-30 (last edited, not re-checked)