Baidu's open-source end-to-end long document OCR model, Unlimited-OCR

Baidu has open-sourced its new OCR model, Unlimited OCR, designed for parsing documents dozens of pages long in a single pass. It achieved a state-of-the-art score of 93.23% on OmniDocBench, surpassing DeepSeek OCR. The model's core innovation is the Reference Sliding Window Attention (R-SWA) mechanism, which uses a "soft forgetting" strategy to maintain a constant key-value cache, ensuring inference speed does not increase with document length. At 6000 tokens, the TPS (transactions per second) is improved by approximately 35%.