O
olmOCR 2
Description
olmOCR 2 is an open model with about 8 billion parameters that turns a document page image into text. It is fine-tuned from Qwen2.5-VL-7B-Instruct and then further trained with reinforcement learning on math equations, tables and other hard cases. The input is one page image rendered with the longest side of 1288 pixels, plus a prompt with page metadata; the authors recommend their olmOCR toolkit for building it. The page reports 82.3 on the olmOCR-Bench benchmark. This is the full-precision version; the authors suggest the FP8 version for most uses. The weights are under Apache 2.0 (for research and educational use under the authors' responsible use guidelines) and are not gated. Hardware requirements are not stated.
Pros and cons
Pros
- Apache 2.0 license
- Tuned for equations, tables and other hard pages
- Open weights without an access request
- Comes with a toolkit for building the input
Cons
- Large model, hardware requirements are not stated
- Works one page at a time
- The license page ties use to the authors' responsible use guidelines
Key facts
- License
- Apache 2.0
- Commercial use
- Allowed
- Developer
- Ai2
- Checked on
- Oct 20, 2026
Editorial score
How strongly the AI Expertise editors recommend this open model for each of its tasks right now, on a 0–10 scale. The score combines output quality, capabilities, ease of launch, license openness and freshness.