Nemotron Nano 12B v2 VL: benchmark results
NVIDIA's Nemotron Nano 12B v2 VL, a document-focused vision-language model pairing a CRadioV2-H vision encoder with the Nemotron-Nano-12B-v2 language model, for multi-image and video understanding. Provider: NVIDIA. Released 2025-10-28. Access: Open.
Aggregate Rating 1508 ± 12, rank #757 of 1635 rated models, from 144 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - story_quality | 1.25 | Dataset z-score | 97.5 |
| AGC-Bench - grapheval_ai_researcher | 0.87 | Dataset z-score | 88.9 |
| AGC-Bench - hypobench | 0.38 | Dataset z-score | 87.8 |
| AGC-Bench - grapheval_iclr | 0.74 | Dataset z-score | 84.1 |
| AA LiveCodeBench | 69.42 | Pass@1 (%) | 78.2 |
| MemLens (32K) - Information Extraction | 63.82 | Accuracy (%) on the information extraction (IE, 246 question | 76 |
| AGC-Bench - puntuguese | 0.58 | Dataset z-score | 75.6 |
| AGC-Bench - thenextchapter | 0.56 | Dataset z-score | 75.6 |
| AGC-Bench - tinystories | 0.62 | Dataset z-score | 75.6 |
| MemLens (64K) - Information Extraction | 56.1 | Accuracy (%) on the information extraction (IE, 246 question | 71.2 |
| AA AIME 2025 | 75 | Accuracy (%) | 70.3 |
| MemLens (32K) - Knowledge Update | 39.66 | Accuracy (%) on the knowledge update (KU, 116 questions) at | 69.2 |
Interactive version: theaggregate.ai/model?slug=nemotron-nano-12b-v2-vl · How It Works · Data refreshed daily, snapshot 2026-11-05.