Sparse Reference Sliding Window Attention (SR-SWA): Compresses text-visual attention interactions from ultra-long sequences down to a fixed-size set of working references (e.g., 512 tokens).
Alternatives to Unlimited-OCR-2
120 open-weight models that do the same job as Unlimited-OCR-2 (3.3B OCR), ranked by how close they are in task and size, then by popularity and how recent they are. Licence and VRAM are shown for each.
MonkeyOCR: Document Parsing with a Structure-Recognition-Relation Triplet Paradigm Zhang Li, Yuliang Liu, Qiang Liu, Zhiyin Ma, Ziyang Zhang, Shuo Zhang, Zidun Guo, Jiarui Zhang, Xinyu Wang, Xiang Bai
?BUnknown licence118 515Jun 2025- Same category
Falcon OCR is a 270M parameter early-fusion vision-language model for document OCR. Given an image, it can produce plain text, LaTeX for formulas, or HTML for tables.
270M16K ctx Apache 2.05.6K 179Feb 2026- Same category
- Smaller (270M vs 3.3B)
- Commercial use OK
PP-OCRv6 is a lightweight OCR system that combines architectural innovation with data-centric optimization. It redesigns the backbone, detection neck, and recognition neck around a unified MetaFormer-style building block with structural reparameterization.
22MApache 2.02.2K 143Jun 2026- Same category
- Smaller (22M vs 3.3B)
- Commercial use OK
PP-OCRv6 is a lightweight OCR system that combines architectural innovation with data-centric optimization. It redesigns the backbone, detection neck, and recognition neck around a unified MetaFormer-style building block with structural reparameterization.
19MApache 2.01.8K 28Jun 2026- Same category
- Smaller (19M vs 3.3B)
- Commercial use OK
PP-OCRv6 is a lightweight OCR system that combines architectural innovation with data-centric optimization. It redesigns the backbone, detection neck, and recognition neck around a unified MetaFormer-style building block with structural reparameterization.
438KApache 2.01.2K 28Jun 2026- Same category
- Smaller (438K vs 3.3B)
- Commercial use OK
PP-OCRv6 is a lightweight OCR system that combines architectural innovation with data-centric optimization. It redesigns the backbone, detection neck, and recognition neck around a unified MetaFormer-style building block with structural reparameterization.
2MApache 2.0716 29Jun 2026- Same category
- Smaller (2M vs 3.3B)
- Commercial use OK
This Hub repository contains a HuggingFace's transformers implementation of Florence-2 model from Microsoft.
232M1K ctx MIT155.6K 155Jun 2024- Same category
- Smaller (232M vs 3.3B)
- Commercial use OK
Dolphin (Document Image Parsing via Heterogeneous Anchor Prompting) is a novel multimodal document image parsing model that follows an analyze-then-parse paradigm.
398M4K ctx MIT522 39Oct 2025- Same category
- Smaller (398M vs 3.3B)
- Commercial use OK
Stockmark-Nemotron-3-Nano-Omni-JapanDocReader is a Japanese document-reading multimodal model built on top of NVIDIA's Nemotron-3-Nano-Omni-30B-A3B-Reasoning. This model injects Japanese structured document parsing capability while preserving as much of the original VQA ability as possible.
33B256K ctx nvidia-open-model-agreement171 8Jul 2026- Same category
- Larger (33B vs 3.3B)
- Commercial use OK
Dolphin (Document Image Parsing via Heterogeneous Anchor Prompting) is a novel multimodal document image parsing model that follows an analyze-then-parse paradigm.
398M4K ctx MIT668 521May 2025- Same category
- Smaller (398M vs 3.3B)
- Commercial use OK
Pix2Struct is an image encoder - text decoder model that is trained on image-text pairs for various tasks, including image captionning and visual question answering. The full list of available models can be found on the Table 1 of the paper:
?BApache 2.0212 39Mar 2023- Same category
- Commercial use OK
Pix2Struct is an image encoder - text decoder model that is trained on image-text pairs for various tasks, including image captionning and visual question answering. The full list of available models can be found on the Table 1 of the paper:
?BApache 2.0179 34Mar 2023- Same category
- Commercial use OK
Pix2Struct is an image encoder - text decoder model that is trained on image-text pairs for various tasks, including image captionning and visual question answering. The full list of available models can be found on the Table 1 of the paper:
?BApache 2.0179 25Mar 2023- Same category
- Commercial use OK
texify is an open-weight ocr model with 312.5M parameters from vikp, published on Hugging Face.
312M2K ctx CC BY-SA 4.047.4K 15Dec 2023- Same category
- Smaller (312M vs 3.3B)
- Commercial use OK
This model is the pretrained version of Pix2Struct, use this model for fine-tuning purposes only.
282MApache 2.010.8K 91Mar 2023- Same category
- Smaller (282M vs 3.3B)
- Commercial use OK
Nougat model trained on PDF-to-markdown. It was introduced in the paper Nougat: Neural Optical Understanding for Academic Documents by Blecher et al. and first released in this repository.
349M4K ctx CC BY-NC 4.02.3K 195Sept 2023- Same category
- Smaller (349M vs 3.3B)
Pix2Struct is an image encoder - text decoder model that is trained on image-text pairs for various tasks, including image captionning and visual question answering. The full list of available models can be found on the Table 1 of the paper:
282MApache 2.03K 36Mar 2023- Same category
- Smaller (282M vs 3.3B)
- Commercial use OK
Nougat model trained on PDF-to-markdown. It was introduced in the paper Nougat: Neural Optical Understanding for Academic Documents by Blecher et al. and first released in this repository.
247M4K ctx CC BY 4.02.4K 35Sept 2023- Same category
- Smaller (247M vs 3.3B)
- Commercial use OK
Pix2Struct is an image encoder - text decoder model that is trained on image-text pairs for various tasks, including image captionning and visual question answering. The full list of available models can be found on the Table 1 of the paper:
282MApache 2.01.9K 50Mar 2023- Same category
- Smaller (282M vs 3.3B)
- Commercial use OK
Pix2Struct is an image encoder - text decoder model that is trained on image-text pairs for various tasks, including image captionning and visual question answering. The full list of available models can be found on the Table 1 of the paper:
282MApache 2.0951 47Mar 2023- Same category
- Smaller (282M vs 3.3B)
- Commercial use OK
Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.
4.7B256K ctx Apache 2.04.4M 1.3KFeb 2026- Related category
- Similar size (4.7B vs 3.3B)
- Commercial use OK
- Widely used
Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.
2.3B256K ctx Apache 2.05.5M 494Feb 2026- Related category
- Similar size (2.3B vs 3.3B)
- Commercial use OK
- Widely used
LFM2.5-VL-3B is a multimodal variant of LFM2.5, a family of hybrid models designed for on-device deployment. It builds on LFM2-VL-3B with further mid- and post-training.
3.1B32K ctx lfm1.045.2K 239Aug 2026- Related category
- Similar size (3.1B vs 3.3B)
Meet Qwen3-VL — the most powerful vision-language model in the Qwen series to date.
4.4B256K ctx Apache 2.02.9M 543Oct 2025- Related category
- Similar size (4.4B vs 3.3B)
- Commercial use OK
- Widely used