What it does
olmOCR is a toolkit for converting PDFs and other image-based document formats into clean, readable plain text. PDF, PNG and JPEG inputs become Markdown, with support described for equations, tables, handwriting and complex formatting, removal of headers and footers, and a natural reading order despite figures and multi-column layouts. It is based on a 7B-parameter vision-language model, so local use needs a recent NVIDIA GPU with at least 12 GB of memory and about 30 GB of free disk. Alternatively, a lightweight install can point at a remote vLLM server or another inference provider that offers an OpenAI-compatible API. Docker images and an AWS S3 work queue for spreading millions of PDFs across several machines are also described, and the README puts the cost at under 200 USD per million pages converted. The project ships olmOCR-Bench, a benchmark of over 7,000 test cases across 1,400 documents, and links an online demo. Its news list records a series of model releases, the latest being v0.4.0 in October 2025.
Capabilities
- Converts PDF, PNG and JPEG documents to Markdown
- Handles equations, tables, handwriting and multi-column layouts
- Removes headers and footers and keeps a natural reading order
- Runs on a local GPU, a remote vLLM server or in Docker
- Multi-node batch processing through an AWS S3 work queue
- olmOCR-Bench benchmark suite for comparing OCR systems
When to use it
- Turning large collections of PDFs into text for LLM datasets or training
- Extracting readable Markdown from scanned or complex documents
- Comparing OCR systems with a ready-made benchmark
- Trying the conversion first through the online demo
