o
AI Tool Profile

olmOCR

Open-source toolkit from Ai2 that converts PDFs and image-based documents into clean plain text or Markdown, for developers. Free under Apache 2.0; local use needs an NVIDIA GPU with at least 12 GB of RAM.

Website
github.com
Pricing model
Open Source
Price start
Free

Description of olmOCR

olmOCR is an open-source toolkit in the allenai GitHub organization for converting PDFs and other image-based document formats into clean, readable plain text. According to the README, it converts PDF, PNG and JPEG files to Markdown and handles equations, tables and handwriting, removes headers and footers, and aims for a natural reading order. It is aimed at developers who run document conversion themselves.

For local inference, the README says a recent NVIDIA GPU with at least 12 GB of GPU RAM and 30 GB of free disk space is needed, and lists RTX 4090, L40S, A100 and H100 cards as tested. Alternatively, a lightweight package can be used with a remote vLLM or OpenAI-compatible server, which avoids needing a GPU on your own machine. The latest release listed is v0.4.0 from October 21, 2025, which introduced the olmOCR-2-7B-1025-FP8 model.

The project states it is efficient at less than $200 USD per million pages converted, which is the maintainers' own claim. The README also lists per-token prices from external providers serving olmOCR-2-7B-1025-FP8, including Cirrascale, DeepInfra and Parasail, and links to an online demo at olmocr.allenai.org. The toolkit has no hosted API of its own. It is licensed under Apache 2.0.

Features of olmOCR

  1. Converts PDF, PNG and JPEG files to Markdown, handling equations, tables and handwriting.
  2. Removes headers and footers and aims for a natural reading order.
  3. Runs locally on an NVIDIA GPU with 12 GB or more of RAM, or against a remote vLLM or OpenAI-compatible server.

Frequently asked questions

Is olmOCR free?

The software is licensed under Apache 2.0 and is free to use. Running it locally needs an NVIDIA GPU with at least 12 GB of RAM; third-party providers that serve the model charge per token.

Does olmOCR have a hosted API?

The README says it does not offer its own hosted API. It links to an online demo and lists external providers that can serve the model through an OpenAI-compatible endpoint.

Website: https://github.com/allenai/olmocr

Company: Ai2 (Allen Institute for AI)

Key facts

Open-source PDF and image to Markdown toolkit from Ai2, for developers with GPUs or an inference provider. Biggest limit: needs an NVIDIA GPU with 12 GB or more, or a third-party server. Free under Apache 2.0.

Pricing model Open source
Free tier Yes · Free open-source software (Apache 2.0); local use needs NVIDIA GPU with 12 GB or more RAM · no credit card required
Starting price Not verified
Data residency On-premises
Compliance (as stated by vendor) Not verified
Integrations vLLM, OpenAI-compatible servers
Public API No
Self-hostable Yes
Official site Not verified
Can a general AI assistant replace it? For no-GPU use, compare Mistral OCR or Google Document AI; for CPU-friendly local parsing, Docling or Tesseract.
Last verified 2026-11-17 by editor

Alternatives & Similar Tools

Vidsembly logo
Vidsembly Freemium

Vidsembly turns PowerPoint decks, PDFs and documents into narrated videos with AI voiceover or your own voice clone in 12 languages, then lets you host and share them with a link or embed.

A

Microsoft Azure service that extracts text, key-value pairs and tables from printed and handwritten forms, PDFs and images, for developers. Microsoft says free and standard pricing options exist; limits are on a page not readable today.

T
Tesseract OCR Open Source

Open-source OCR engine and command-line program that recognizes text in images, for developers building OCR into their own software. Free under the Apache License 2.0 and recognizes more than 100 languages out of the box.

M

Mistral AI's OCR API that extracts text and structured content from PDFs and images, for developers building document pipelines. Mistral lists OCR 4.1 at $4 per 1,000 pages; no free-tier terms were stated on the pages read.

D
Docling Open Source

Open-source toolkit that parses PDFs, Office files, images and other formats into structured output for generative AI pipelines, for developers. Free under the MIT license and can run locally, including air-gapped.

N
Numerous.ai Freemium

Numerous.ai brings ChatGPT into Google Sheets and Microsoft Excel through an =AI function you can use in any cell, with no API key needed. Its site does not publish plan prices, so check the add-on before you rely on it.

Update history

  1. 2026-11-17 · ai alternative note updated (editor)
  2. 2026-11-17 · editor verdict updated (editor)
  3. 2026-11-17 · self hostable updated (editor)
  4. 2026-11-17 · api available updated (editor)
  5. 2026-11-17 · integrations updated (editor)
  6. 2026-11-17 · data residency updated (editor)
  7. 2026-11-17 · free tier updated (editor)
  8. 2026-11-17 · pricing model updated (editor)

Beyond the tool

Need the outcome rather than the software?