Radar ·
OCR model benchmark under 1B parameters on OmniParseBench
A public post reports that a team evaluated recent small OCR models, under 1B parameters, on OmniParseBench. The post gives no link to a primary source or results page.
Radar records a public post stating that the author's team evaluated recent small OCR models, under 1B parameters, on OmniParseBench. According to the text, the benchmark serves as a reference point for choosing small OCR models. The post reports no metrics, document counts, model versions or performance conclusions, and it links to neither a primary source nor a results page, so the claims could not be confirmed from the material itself.
To consult the original, look for the OmniParseBench publication and the official documentation of the models cited, checking dates, license, methodology and results tables before using any figure. If you study the topic with AI assistance, do not paste real documents containing personal data; use synthetic or anonymized samples. Rota Nacional detects personal data before any model runs, but deciding what is sent remains your organization's responsibility. OCR jobs are in preview, while document extraction already passes through the same protection barrier.
Original source: Benchmark de modelos OCR com menos de 1B de parâmetros no OmniParseBench