Nanonets OCR

license

none-declared(the Hugging Face model card returns no license field in its metadata, so the weights are published without a declared grant)

weights

open(downloadable from Hugging Face without a gate)

data

closed(no training corpus is published or described)

code

closed(no training or fine-tuning code is published

Nanonets publishes the weights on Hugging Face without a gate, but the model card declares no license, and no training data or training code accompany the release. With no license there is nothing that permits using the weights, so the release scores 2, above closed weights and below any licensed open-weight model.

• https://huggingface.co/api/models/nanonets/Nanonets-OCR2-3B recorded 2026-09-16

  Hugging Face model API record for nanonets/Nanonets-OCR2-3B: 38,340 downloads in the trailing 30 days, 516 likes, weights downloadable, and NO license field in cardData. Establishes that the weights are open, that no licence is declared where the platform expects one, and that no training data or code accompanies the release.

Adoption is measured as Hugging Face downloads of the Nanonets-OCR2-3B checkpoint over the trailing month, the only usage signal the release has. That reflects interest in the open-weight model rather than use of the hosted platform Nanonets sells alongside it.

• https://huggingface.co/api/models/nanonets/Nanonets-OCR2-3B recorded 2026-09-16

  Hugging Face model API record: 38,340 downloads in the trailing 30 days.

Nanonets OCR preserves document structure further than most peers: signatures, checkboxes and watermarks come out as tagged elements instead of flattened text, matching Chandra's output. What it lacks is customer-configurable entity types; its tags are a fixed set the model defines, not a schema a customer can set.

• https://huggingface.co/api/models/nanonets/Nanonets-OCR2-3B recorded 2026-09-16

  The model's Hugging Face record, establishing what is published and its documented output structure.