Radar ·
Unlimited-OCR-2 tested on multi-page PDFs on Mac and DGX Spark
An engineer ran an open source document parsing model with 3.3 billion parameters on a 6-page PDF across three machines and checked the output against the PDF's own text.
According to the report published on October 27, 2026, an engineer ran Unlimited-OCR-2, an open source document parsing model with 3.3 billion parameters, on the same 6-page PDF across three machines. The output was verified against the PDF's own text. The post reports throughput figures and says the base mode got one page wrong when requests reached 4, a detail the source shows truncated.
The material is relevant to anyone evaluating OCR for long documents, running locally or on a GPU server. The available excerpt does not give the full methodology, such as versions, hardware configuration or repetitions, so the numbers should be read as a single case. To check, open the original post and compare the metric and test conditions. At Rota Nacional, document extraction already passes through the same personal data barrier before any model runs, but OCR is still in preview.
Original source: Unlimited-OCR-2 avaliado em PDFs de várias páginas em Mac e DGX Spark