NVIDIANemotron 3.5
Nemotron 3.5 Parse
Parse converts page images into document content and structure. Its outputs include text, tables, chart data, reading order and bounding boxes.
What the release provides
The 0.631B-parameter model exports Markdown, HTML or structured JSON. The repository offers single-, double- and triple-pass parsing, plus an optional 89.3M-parameter DFlash2 draft model.Nemotron 3.5 Parse official model card
How to use it well
Practical guidance. Use structured output when table cells or reading order matter. Retain the original page and coordinates for review. Compare parsing modes on difficult pages; a higher benchmark score does not prove that a particular scan or handwritten form is correct.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | nvidia/Nemotron-3.5-Parse |