Baidu's Unlimited OCR Model Tops Four Charts on GitHub & HuggingFace, Achieves 93.92% Accuracy in Document Parsing
Baidu has open-sourced its Unlimited OCR model for long document parsing. It quickly topped four trending charts on GitHub and HuggingFace. The model achieved a 93.92% score on the OmniDocBench benchmark, setting a new record for end-to-end OCR performance. With 3B total parameters but only about 570M activated during inference, it balances power and efficiency.
Read More