LightOn OCR-3
An open-weight OCR model released by LightOn, designed for document intelligence scenarios. It supports processing of scanned documents, PDFs, and tabular documents, with strong capabilities in text localization and layout reconstruction.
Timeline (last 90 days)
LightOn OCR-3 is now available on the OpenDocRouter platform.
LightOn OCR-3 is priced at $0.28 per million input tokens and $1.40 per million output tokens.
LightOn OCR-3 costs approximately $3.19 per thousand pages of documents (based on ParseBench calculations).
LightOn OCR-3 is an open-weight OCR model.
According to the official ParseBench benchmark, LightOn OCR-3 sits on the Pareto frontier of open-weight OCR models.
LightOn OCR-3 performs comparably to Gemini 3.8 flash low but is priced approximately 45% lower.
LightOn OCR-3 delivers decent performance on table extraction.
LightOn OCR-3 performs quite good on localization/grounding (spatial text position recognition).
Open-weight models theoretically allow teams to deploy the same set of weights in their own environments, satisfying compliance requirements for data residency.
It is recommended to use OCR-3 for main body text extraction where localization requirements are high and cost sensitivity is a concern, while retaining manual review or higher-tier models as a fallback for chart-intensive sections or cases requiring extreme precision.
1 more timeline events
All Facts (11)
The significantly higher pricing of OCR model output tokens compared to input tokens is consistent with the general pattern of generative models.
50%UnverifiedLightOn OCR-3 is priced at $0.28 per million input tokens and $1.40 per million output tokens.
50%UnverifiedLightOn OCR-3 costs approximately $3.19 per thousand pages of documents (based on ParseBench calculations).
50%UnverifiedLightOn OCR-3 is an open-weight OCR model.
50%UnverifiedAccording to the official ParseBench benchmark, LightOn OCR-3 sits on the Pareto frontier of open-weight OCR models.
50%UnverifiedLightOn OCR-3 performs comparably to Gemini 3.8 flash low but is priced approximately 45% lower.
50%UnverifiedLightOn OCR-3 delivers decent performance on table extraction.
50%UnverifiedLightOn OCR-3 performs quite good on localization/grounding (spatial text position recognition).
50%UnverifiedOpen-weight models theoretically allow teams to deploy the same set of weights in their own environments, satisfying compliance requirements for data residency.
50%UnverifiedIt is recommended to use OCR-3 for main body text extraction where localization requirements are high and cost sensitivity is a concern, while retaining manual review or higher-tier models as a fallback for chart-intensive sections or cases requiring extreme precision.
50%UnverifiedLightOn OCR-3 is now available on the OpenDocRouter platform.
50%