NVIDIA releases Nemotron Parse 2.0: chart-aware parsing and multilingual OCR

pmttyji · reddit · 2026-08-06

NVIDIA released Nemotron Parse 2.0 on Hugging Face, which converts document images into structured representations with text, layout classes, bounding boxes, and reading order. Compared to v1.2, it adds 20k token vocabulary expansion for multilingual support, chart-aware parsing with <classChart>, and improved table handling. It's designed for document understanding, RAG, data extraction, and more.

Original post →

More from Multimodal

Multimodal channel →