selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

Alternatives to OvisOCR3

120 open-weight models that do the same job as OvisOCR3 (853M 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.

the original
ATH-MaaS logo

OvisOCR3

ATH-MaaS·OCR

≈2.4 GB

OvisOCR3 is a unified 0.8B multimodal model for OCR and document understanding. It integrates document parsing, general OCR, key information extraction, document figure captioning, and cross-page table reconstruction into a single compact model.

853M256K ctx Apache 2.0
437.7KOct 2026gguf
--vram
any
--gpus
--quant
  1. facebook logo

    nougat-small

    facebook·OCR

    ≈1.1 GB

    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 4.0
    2.4KSept 2023
    • Same category
    • Smaller (247M vs 853M)
    • Commercial use OK
    vs
  2. nvidia logo

    nemotron-ocr-v1

    nvidia·OCR

    ? GB

    The Nemotron OCR v1 model is a state-of-the-art text recognition model designed for robust end-to-end optical character recognition (OCR) on complex real-world images.

    ?BNVIDIA Open Model License
    232Oct 2025
    • Same category
    • Commercial use OK
    vs
  3. allenai logo

    olmOCR-2-7B-1025

    allenai·OCR

    ≈10 GB

    Full BF16 version of olmOCR-2-7B-1025-FP8. We recommend using the FP8 version for all practical purposes except further fine tuning.

    8.3B125K ctx Apache 2.0
    82.8KOct 2025gguf
    • Same category
    • Larger (8.3B vs 853M)
    • Commercial use OK
    vs
  4. datalab-to logo

    indra-ocr

    datalab-to·OCR

    ≈0.5 GB

    Indra OCR is a 1M-parameter OCR model that reads up to 49 pages per second on CPU. It turns PDFs and images into markdown, plain text, or JSON with boxes, and it has these features:

    1MOpenRAIL
    6KOct 2026
    • Same category
    • Smaller (1M vs 853M)
    vs
  5. PaddlePaddle logo

    PP-OCRv6_small_rec

    PaddlePaddle·OCR

    ≈0.5 GB

    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.

    5MApache 2.0
    12.3KJun 2026
    • Same category
    • Smaller (5M vs 853M)
    • Commercial use OK
    vs
  6. echo840 logo

    MonkeyOCR

    echo840·OCR

    ? GB

    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 licence
    116Jun 2025
    • Same category
    vs
  7. PaddlePaddle logo≈0.6 GB

    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 2.0
    2.2KJun 2026
    • Same category
    • Smaller (22M vs 853M)
    • Commercial use OK
    vs
  8. PaddlePaddle logo≈0.5 GB

    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 2.0
    1.9KJun 2026
    • Same category
    • Smaller (19M vs 853M)
    • Commercial use OK
    vs
  9. PaddlePaddle logo≈0.5 GB

    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 2.0
    1.2KJun 2026
    • Same category
    • Smaller (438K vs 853M)
    • Commercial use OK
    vs
  10. allenai logo≈10 GB

    This is a preview release of the olmOCR model that's fine tuned from Qwen2-VL-7B-Instruct using the olmOCR-mix-0225 dataset.

    8.3B32K ctx Apache 2.0
    33.3KJan 2025gguf
    • Same category
    • Larger (8.3B vs 853M)
    • Commercial use OK
    vs
  11. PaddlePaddle logo≈0.5 GB

    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 2.0
    723Jun 2026
    • Same category
    • Smaller (2M vs 853M)
    • Commercial use OK
    vs
  12. datalab-to logo

    chandra

    datalab-to·OCR

    ≈11 GB

    Chandra is an OCR model that outputs markdown, HTML, and JSON. It is highly accurate at extracting text from images and PDFs, while preserving layout information.

    8.8B256K ctx OpenRAIL
    1.8KOct 2025gguf
    • Same category
    • Larger (8.8B vs 853M)
    vs
  13. inclusionAI logo

    ArmorOCR

    inclusionAI·OCR

    ≈11 GB

    ArmorOCR is a two-stage framework for grounded adversarial OCR perception built on Qwen3-VL-8B-Instruct. It enables single-pass inference on the original image, without any inference-time visual transformations or tool assistance.

    8.8B256K ctx Apache 2.0
    149Aug 2026gguf
    • Same category
    • Larger (8.8B vs 853M)
    • Commercial use OK
    vs
  14. 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-agreement
    164Jul 2026gguf
    • Same category
    • Larger (33B vs 853M)
    • Commercial use OK
    vs
  15. 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 2.0
    217Mar 2023
    • Same category
    • Commercial use OK
    vs
  16. allenai logo

    olmOCR-7B-0825

    allenai·OCR

    ≈10 GB

    This is a release of the olmOCR model that's fine tuned from Qwen2.5-VL-7B-Instruct using the olmOCR-mix-0225 dataset.

    8.3B125K ctx Apache 2.0
    669Aug 2025gguf
    • Same category
    • Larger (8.3B vs 853M)
    • Commercial use OK
    vs
  17. allenai logo

    olmOCR-7B-0725

    allenai·OCR

    ≈10 GB

    This is a release of the olmOCR model that's fine tuned from Qwen2.5-VL-7B-Instruct using the olmOCR-mix-0225 dataset.

    8.3B125K ctx Apache 2.0
    771Jul 2025gguf
    • Same category
    • Larger (8.3B vs 853M)
    • Commercial use OK
    vs
  18. 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 2.0
    179Mar 2023
    • Same category
    • Commercial use OK
    vs
  19. 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 2.0
    172Mar 2023
    • Same category
    • Commercial use OK
    vs
  20. reducto logo

    RolmOCR

    reducto·OCR

    ≈10 GB

    Earlier this year, the Allen Institute for AI released olmOCR, an open-source tool that performs document OCR using the Qwen2-VL-7B vision language model (VLM).

    8.3B125K ctx Apache 2.0
    1.4KApr 2025gguf
    • Same category
    • Larger (8.3B vs 853M)
    • Commercial use OK
    vs
  21. Qwen logo

    Qwen3.5-0.8B

    Qwen·Vision

    ≈1.4 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    873M256K ctx Apache 2.0
    2.5MFeb 2026ollama
    • Related category
    • Similar size (873M vs 853M)
    • Commercial use OK
    • Widely used
    vs
  22. nvidia logo≈1.4 GB

    NVIDIA Nemotron Parse 2.0 transforms document images into structured, machine-readable representations with text, layout classes, bounding boxes, and reading-order information.

    903M9K ctx Custom licence
    650.6KJun 2026
    • Related category
    • Similar size (903M vs 853M)
    • Commercial use OK
    vs
  23. Qwen logo

    Qwen3.5-0.8B-Base

    Qwen·Vision

    ≈1.4 GB

    Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

    873M256K ctx Apache 2.0
    880.7KFeb 2026gguf
    • Related category
    • Similar size (873M vs 853M)
    • Commercial use OK
    vs
  24. internlm logo

    Intern-Decision-0.8B

    internlm·Vision

    ≈1.4 GB

    Intern-Decision-0.8B is a multimodal structured decision model fine-tuned from Qwen3.5-0.8B. It accepts a shared state, a schema of named questions, and optional images, and returns an answer distribution for every question in one model forward pass.

    853M256K ctx Apache 2.0
    4.1KSept 2026gguf
    • Related category
    • Similar size (853M vs 853M)
    • Commercial use OK
    vs