Baidu released and open-sourced Unlimited OCR, an end-to-end model designed for
long-document parsing. The day after release it ranked first on GitHub Daily
Trending and the GitHub Python chart and led Hugging Face’s global model and
multimodal model trend lists. Unlimited OCR has 3 billion parameters and uses
roughly 570 million active parameters during inference. In public testing it
scored 93.92% on the OmniDocBench v1.6 benchmark, setting a new end-to-end OCR
record.