- Released
- 1y ago
- Size
- 8.3Bparameters
Ai2 open vision-language model, fine-tuned from Qwen2-VL 7B, for turning PDFs into clean plain text.
How it compares
Frequently asked questions
Running locally
olmOCR is an 8.3B model with estimated VRAM needs of 6.3 GB at Q4, 10.6 GB at Q8 and 19.6 GB at FP16. These figures are estimates, weights plus about 15% and 0.5 GB for a short context.
Yes. Based on estimates, it fits a 24 GB GPU at FP16 (19.6 GB), a 16 GB or 12 GB GPU at Q8 (10.6 GB) and an 8 GB GPU at Q4 (6.3 GB).
Yes. At Q4, with an estimated 6.3 GB, it fits the smallest common Mac with 16 GB of memory, assuming macOS lets the GPU use about 75% of unified memory.
Privacy and data
Yes, olmOCR has open weights, so you can self-host it for full control of your data.
Capabilities and specs
It handles text and image inputs, but not audio or video.
Quality and benchmarks
The closest alternatives, matched on model kind and release date with no comparable price, are Phi-4-mini by Microsoft, Phi-4-multimodal-instruct by Microsoft and Command R7B Arabic by Cohere.
Phi-4-miniolmOCR vs Phi-4-miniPhi-4-multimodal-instructolmOCR vs Phi-4-multimodal-instructCommand R7B ArabicolmOCR vs Command R7B Arabic
Licensing and access
olmOCR has open weights, released under the Apache 2.0 license.
Yes, the Apache 2.0 license is permissive and allows commercial use.
You can download the weights from the Hugging Face repository allenai/olmOCR-7B-0225-preview.
Basics
olmOCR was made by Ai2 and released on February 25, 2025.
Yes, olmOCR has open weights.
Generated from Vertical Tabs data, updated November 3, 2026.
Discussion
No comments yet
Be the first to share your thoughts.