LightOnOCR-3-0.8B

Use it to transcribe documents and extract visual information such as labeled regions and chart data.

Availability

Model linked from the LightOnOCR-3 model card

Pricing model

Not established

Deployment

Transformers or vLLM usage instructions are provided for the model family

Privacy

Not established

Licence

Apache License 2.0 for research and commercial use

• In transcription mode, the model outputs page text as markdown. ^[1]

  * 📝 **Transcription mode:** call the model with an empty prompt and get the full page text, as before.

• In grounding mode, the model returns content blocks with labels and bounding boxes. ^[1]

  * 📍 **Grounding mode:** call it with the `grounding` prompt and every block comes back with a label and a bounding box

• Image blocks include short descriptions, and chart blocks contain tables of extracted data points. ^[1]

  * **Chart blocks** contain an HTML table of the data points extracted from the figure, turning visual information into structured data.

• The model can be loaded with Transformers or served with vLLM. ^[1]
• The 0.8B variant is described as fast and efficient. ^[2]

  The 1B keeps the LightOnOCR-2-1B architecture, while the 0.8B and 4B adopt the Qwen3.5 vision-language architecture, which simplifies integration with existing tools and brings a significant speed-up.

• LightOnOCR-3 models provide bounding boxes and labels for visual document elements.
• LightOnOCR-3 models can provide image descriptions and extract chart data into tables.
• The LightOnOCR-3-0.8B model is linked from the model card.
• The model family includes usage instructions for Transformers and vLLM.
• The model family is released under Apache License 2.0 for research and commercial use. ^[2]

  All models are released under the Apache 2.0 license for research and commercial use.

• Extract document layout and chart data, not just text Daily brief 2026-10-08