Nemotron 3.5 Super VL
nvidia · Released Oct 2026
Nemotron 3.5 Super VL is NVIDIA's open-weight 120B vision-language model (12B active) under the OpenMDW-1.1 licence, with a 1M-token context and text, image and video input. The deciding caveat is hardware and evidence: NVIDIA recommends 4x H100 or 2x GB200 for the BF16 build, and the card's scores are NVIDIA's own runs. Suits teams that can host a large model on their own GPUs and want open weights for document and visual work.
Is it right for you?
Good for
- Self-hosted document and visual work: open weights, text/image/video input, OCR, tables and charts per NVIDIA
- Teams wanting a permissive licence: OpenMDW-1.1 is stated as ready for commercial or non-commercial use
- Flexible deployment: BF16 and NVFP4 checkpoints, plus NVIDIA NIM and OpenRouter availability
Not good for
- Small GPU budgets: the BF16 card recommends 4x H100 or 2x GB200, and the NVFP4 card lists 1x DGX Spark (text and images), 1x H200 or 2x H100, or 1x GB200
- Independent evidence: card scores are collected by NVIDIA via NeMo Gym, and the pages we read show no third-party run
How it performs by task
Reasoning
GPQA-Diamond 89.52 on the model card (NVIDIA run)
Coding
LiveCodeBench v6 88.22 and SWE-bench Verified 71.10 on the model card (NVIDIA run)
Document and visual understanding
NVIDIA lists OCR, tables, charts and structured information in documents; no document-specific benchmark read
Pricing
Input
Not stated
Output
Not stated
Context
Open weights under OpenMDW-1.1; 1M ctx; hosted NIM and OpenRouter prices not verified
Benchmarks
No verdict changes yet
The clock starts day one. Changes land here as our verdict evolves.
Sources
Verification log
No verification checks yet
We haven't logged a verification check for this entry. Once a check runs, its history shows here.