A compact 155M-parameter OCR model mapping images directly to transcribed text. Supports Japanese, Korean, Chinese, and English recognition.