DeepSeek's MoE model with 552B total parameters and 8-16B active per token. 1M-token context, MIT license: free commercial use. Built for online serving, too large for local use on a single PC.
NVIDIA Nemotron Parse (famiglia OCR)
Released on Jun 30, 2026
NVIDIA's vision encoder-decoder family for local document understanding: markdown, LaTeX formulas, bounding boxes and a semantic class per element, in reading order. ViT-H (C-RADIO) + 10-block mBART decoder, all under 1 billion parameters. Locally via vLLM (v0.20–v0.26 for 2.0) or Transformers, on NVIDIA GPUs (Turing through Blackwell).
Features
- ViT-H encoder + 10-block mBART decoder
- Under 1 billion parameters (≈3.6–3.8 GB in bfloat16)
- Markdown, LaTeX, bounding boxes and semantic classes
- Locally via vLLM v0.20–v0.26 (2.0) or Transformers
- NVIDIA GPUs only (Turing, Ampere, Hopper, Blackwell)
Version history
- 2.0Latest
Versione attuale (~903M di parametri): OCR multilingua esteso con grandi guadagni su CJK e scritture indiche (IndicVisionBench ANLS 0.06→0.72), miglioramento del manoscritto, patch per i pesi legati in vLLM. Supporta vLLM v0.20–v0.26, validato con v0.20.0.
- v1.2
Miglioramento dell'estrazione su pagine miste e scrittura informale; disponibile in vllm main (immagine vllm-openai v0.14.1).
- v1.1
Rilascio iniziale (~957M di parametri): markdown, LaTeX, bounding box e classi semantiche nell'ordine di lettura; paper arXiv 2511.20478. In locale via vLLM (immagine vllm-openai v0.14.1) e validato anche con vLLM 0.20.1.
Tags
Useful links
Discover NVIDIA Nemotron Parse (famiglia OCR)
Visit the official site to download, learn more and start using NVIDIA Nemotron Parse (famiglia OCR).
Related resources — Models
The most advanced open image model: excellent prompt adherence, solid anatomy, full fine-tuning. The standard for serious work.
Very few steps, high quality, and strong text rendering. The distilled FLUX variant, built for speed.

