olmOCR 2

Name: olmOCR 2
Author: Ai2

Ai2 · released 2025-10-06 · data checked 2026-11-06

olmOCR 2 is a 8.3B-parameter OCR and document-parsing model from Ai2 (US), released 2025-10-06 under the Apache-2.0 license. It scores 82.4 on olmOCR-Bench (#12 of 24). It needs about 17.6 GB of memory at 16-bit.

Fine-tuned from Qwen2.5-VL 7B by the team behind olmOCR-Bench, with fully open training data.

OmniDocBench v1.6

—

Not reported

olmOCR-Bench

82.4

#12 of 24 models

Parameters

8.3B

OCR specialist

Memory to run

~17.6 GB

16-bit weights plus about 1 GB

Benchmarks

olmOCR-Bench Lab-reported82.4

Best: OXR 1.0 87.8

The tick marks the best score on each benchmark. See how to read the ranking.

Specifications

Lab: Ai2 (US)
Type: OCR / document parsing specialist
Total parameters: 8.3B
License: Apache-2.0
Memory to run: ~17.6 GB at 16-bit
Released: 2025-10-06

Weights and usage on Hugging Face ↗

Closest alternatives

Falcon OCR 1.5 (270M, 82.4 on olmOCR-Bench, Apache-2.0)
MiniCPM-OCR 1.0 (1.3B, 82.7 on olmOCR-Bench, Apache-2.0)
LightOnOCR-2 1B (1B, 83.2 on olmOCR-Bench, Apache-2.0)
jina-ocr-v1 (3.4B, 83.4 on olmOCR-Bench, CC-BY-NC-4.0)

Sources

https://huggingface.co/allenai/olmOCR-2-7B-1025

Something wrong here? Report it.