A specialized text encoder trained specifically for OCR-related tasks, this model excels at understanding noisy, fragmented, or irregularly formatted text that typically comes from document scanning pipelines. It produces embeddings tuned for the kinds of linguistic patterns found in recognized text rather than clean prose. Its narrow focus means it may underperform on general NLP tasks outside the OCR domain.