SSyncDev

Nemotron 3.5 Parse

SyncDev model directory · Profile updated Nov 18, 2026

Compare with another model

Nemotron 3.5 Parse is NVIDIA's 0.631B open document parser that turns page images into Markdown, HTML or JSON with tables, charts and reading order.

Key specifications

Specification score
9.6
License
OpenMDW-1.1

The specification score summarizes recorded features and limits. It is not a benchmark result.

Pricing conditions: Self-hosted weights carry no fee. Hosted NVIDIA Build and Baseten rates are not published on the card, so check your own account.

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Image input
  • Text input

Produces

  • Text output

Capabilities

VisionOpen weights

What is Nemotron 3.5 Parse?

Turning a scanned contract or a slide deck into clean, structured text usually means chaining an OCR engine, a layout detector and a table model. Nemotron 3.5 Parse folds that job into one 0.631B-parameter vision-language model. Give it a page image and it returns text, tables, chart data, reading order, semantic classes and bounding boxes, exportable as Markdown, HTML or structured JSON.

It is small enough that throughput, not accuracy, is usually the constraint. That is why the passes matter.

Choosing a pass mode

Mode What happens Publisher-reported ParseBench overall Pages/s with DFlash2 (one H200)
Single Whole page parsed once 72.74 18.57
Double Tables and charts re-parsed from crops 74.92 14.60
Triple Separate layout detection, then refine the union of regions 75.20 13.49

Single pass is the place to start for bulk ingestion. Double pass buys most of the accuracy gain for table-heavy material; the third pass adds little on top. Since 4 November 2026 the repository also carries an 89.3M-parameter DFlash2 drafter for speculative decoding in vLLM, which speeds generation without changing what gets parsed.

Practical details

  • Input is images, not PDFs. Render each page first (PyMuPDF works). The bundled helper does not stitch tables that continue across pages or build a document-wide reading order.
  • Prompts are control tokens. Bounding boxes, semantic classes and text formatting are switched on or off with reserved tokens rather than a free-form instruction. Keep their order exactly as documented.
  • Overlapping hypotheses. By default the crop merger can keep differing full-page and crop versions of a table. A guarded merge policy picks one, at a small accuracy cost.
  • Languages. Multilingual OCR was evaluated across 163 language/script entries in 26 script groups, plus 10 Indic languages.

Weights are under the OpenMDW-1.1 licence. NVIDIA also exposes the model through its Build API, and a Baseten deployment is offered; check those consoles for current pricing, which changes.

Will it fit your pipeline?

Take fifty pages that broke your current OCR stack: merged cells, rotated charts, two-column layouts. Run them in single and double pass, diff the Markdown against hand-corrected output, and time both. If double pass fixes your tables, the roughly 20% throughput hit is usually worth it.

For a hosted alternative with no GPU to manage, compare it against Mistral OCR 4.1. The rest of NVIDIA's open line-up sits on the NVIDIA provider page, and more parsers are grouped under document OCR models.

How to evaluate Nemotron 3.5 Parse

Compare the model on the work you need to complete.

Tasks and inputs

Nemotron 3.5 Parse is listed for Image-to-text, Document OCR. It accepts image, text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.

Choosing a deployment

The weights are downloadable under OpenMDW-1.1. Check runtime support and size the complete model and working context for your expected concurrency.

Frequently asked questions

Text, tables, chart data, reading order, semantic classes and bounding boxes from a page image, exported as Markdown, HTML or JSON.

These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.

OpenBMB

MiniCPM-OCR 1.0

MiniCPM-OCR 1.0 is OpenBMB's open 1.3B document-parsing model for text, tables, formulas and charts, with a free API and on-device apps.

— context

OpenBMB

MiniCPM-OCR Jev

MiniCPM-OCR Jev is OpenBMB's open 2B judge that scores whether an OCR result for a text, formula or table region is correct, from 0 to 1.

— context

Tencent

Youtu-Parsing-Omni

Youtu-Parsing-Omni is Tencent Youtu's 5B open parser that turns documents, charts, images, audio and video into one structured JSON format.

— context

lightonai

LightOnOCR-3-0.8B

LightOnOCR-3-0.8B is LightOn's smallest open document model: page images in, markdown out, with optional bounding boxes and chart tables.

— context

Sarvam

Sarvam Vision 2.1

Sarvam Vision 2.1 is Sarvam's 3B document-intelligence VLM for OCR, tables, forms and Indic handwriting in 22 Indian languages plus English.

— context

Cohere

Cohere Parse

Cohere Parse (parse-v5.0) is a 2.3B vision-language document parser that turns PDFs, slides and scans into Markdown for RAG, at $1.50 per 1,000 pages.

8.19K context