olmOCR vs Florence-2
Pick olmOCR to turn full PDFs into Markdown at scale, especially academic papers with equations, tables or mixed printed and handwritten pages. Pick Florence-2 for short English text in images where you need boxes per line, on small hardware or in a pipeline that also needs detection. The main split is output: olmOCR returns structured Markdown from a 7B model that needs a 12 GB GPU, while Florence-2 returns only text and boxes from a 0.23B or 0.77B model.
olmOCRallenai/olmocr | Florence-2microsoft/florence-2 | |
|---|---|---|
| Type | Open source | Open source |
| Price | 2 credits / page | 1 credit / page |
| Per 1,000 pages | $2.00 | $1.00 |
| Typical speed | 1.5 s / page | 0.8 s / page |
| Best for | Bulk PDF to text for datasets | Lightweight OCR with boxes |
| Inputs | PDF, images | images |
| Languages | Mainly English | English |
| Max pages per request | 1,000 | 500 |
| Tables | ✓ Yes | No |
| Handwriting | ✓ Yes | No |
| Math | ✓ Yes | No |
| Bounding boxes | No | ✓ Yes |
| Confidence | No | No |
| Forms | No | No |
| Multilingual | No | No |
| On OCRRouter | Available | Available |
Prices are what you pay on OCRRouter. Speed is the provider's typical figure, not a benchmark we ran.
Which to choose
Choose olmOCR if
- You need full-page PDF to Markdown for LLM training or RAG, and it scores 82.4 on olmOCR-Bench, above Marker (76.1), MinerU 2.5.4 (75.2) and Mistral OCR (72.0).
- Your documents have equations, tables or handwriting, and you want headers and footers stripped in natural reading order.
- You want a cost figure at volume: under $200 per million pages self-hosted per Ai2, or $0.09 input / $0.19 output per million tokens on DeepInfra for olmOCR-2-7B.
Choose Florence-2 if
- You need text with quadrilateral boxes per line, which OCR_WITH_REGION returns.
- You have limited hardware or want an edge deployment: the model is 0.23B (base) or 0.77B (large) and MIT licensed.
- Your pipeline also needs captioning, detection or grounding from the same model, on English scene text or short labels.
Key differences
| olmOCR | Florence-2 | |
|---|---|---|
| Output formats | Markdown and JSON | Plain text and bounding boxes. No Markdown, tables or LaTeX. |
| Model size | 7B | 0.23B (base), 0.77B (large) |
| Hardware | NVIDIA GPU with at least 12 GB VRAM and 30 GB free disk | Not published |
| Licence | Apache-2.0 (toolkit) | MIT |
| How you run it | Self-host with vLLM on a GPU, or use third-party hosted endpoints | Self-host via Transformers |
| Handwriting and non-English text | Handles handwriting, including mixed printed and handwritten pages | Spelling errors and missing spaces on cursive handwriting and French text. Chinese output mostly wrong, as the vocabulary lacks CJK characters. |
| Known failure | Out of memory on a 12 GB GPU (issue #424), gets stuck on the last few pages (#433), unreliable image bounding box coordinates (#437) | No document structure output. The continued-pretrained checkpoint used only 0.1B samples and might not be well trained. |
Sources
- olmOCR GitHub README
- Issue #424 OOM
- Issue #437 bbox
- Florence-2-large model card
- HF blog OCR test
- HF discussion Chinese very bad
What each one is
olmOCR
Open-source vision model from Ai2 (often searched as olm ocr) that turns PDFs into clean, reading-order text. Latest release: olmOCR 2.
olmOCR detailsFlorence-2
Small Microsoft vision model (0.23B and 0.77B) with OCR and OCR-with-regions tasks that return text with boxes.
Florence-2 detailsTry both on your own file
Run the file once with each model and compare the output.
Use one, fall back to the other
{"model": "allenai/olmocr","fallbacks": ["microsoft/florence-2"],"file_url": "https://example.com/document.pdf"}Questions
More comparisons
olmOCR alternativesFlorence-2 alternativesBest OCR models in 2026