nemotron-3-nano-30B-a3B: benchmark results
NVIDIA's Nemotron-3-Nano-30B-A3B, a hybrid Mamba-2/MoE model (3.5B active of 30B total, 128 experts) with configurable reasoning traces, supporting English, German, Spanish, French, Italian and Japanese. Provider: NVIDIA. Released 2025-12-15. Access: Open.
Aggregate Rating 1535 ± 12, rank #662 of 1656 rated models, from 273 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - unfun_corpus | 4.91 | Dataset z-score | 100 |
| LLM Stats (WMT24++) | 86.2 | Score (%) | 96.3 |
| AGC-Bench - ocw | 1.49 | Dataset z-score | 92.7 |
| AA AIME 2025 | 91 | Accuracy (%) | 92.1 |
| AGC-Bench - thenextchapter | 1.19 | Dataset z-score | 91.5 |
| UnpredictaBench - Code | 18.49 | KS@100 (%): share of the 238 Code problems (distributions in | 89.5 |
| UnpredictaBench - RealWorld | 33.33 | KS@100 (%): share of the 30 RealWorld problems (natural-lang | 86.8 |
| AA IFBench | 71.09 | Accuracy (%) | 86.4 |
| AA LiveCodeBench | 74.07 | Pass@1 (%) | 85.7 |
| CoCoReviewBench - Category Completeness | 82.93 | Category completeness (%): the average number of review cate | 82.4 |
| AGC-Bench - liveideabench | 0.98 | Dataset z-score | 81.7 |
| AGC-Bench - tinystories | 0.81 | Dataset z-score | 81.7 |
Interactive version: theaggregate.ai/model?slug=nemotron-3-nano-30b-a3b · How It Works · Data refreshed daily, snapshot 2026-10-20.