NVIDIA-Nemotron-Nano-9B-v2-Japanese: benchmark results
NVIDIA's Nemotron Nano 9B v2 Japanese, a hybrid Mamba-2/attention model further trained from NVIDIA-Nemotron-Nano-9B-v2 for Japanese, with toggleable reasoning traces controlled by system prompt. Provider: NVIDIA. Released 2026-02-17. Access: Open.
Aggregate Rating 1542 ± 17, rank #634 of 1635 rated models, from 138 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - BFCL - Live AST | 74.07 | Accuracy (%) | 89.9 |
| Nejumi 4 - ALT - Bias | 93.44 | Score (%) | 88.6 |
| Swallow - English MT-Bench - Math | 98.7 | Judge Score (normalized, %) | 81.3 |
| Nejumi 4 - Toxicity - Violation Categories | 48.41 | Criteria met (%) | 78.8 |
| Swallow - Japanese MT-Bench - Math | 97.8 | Judge Score (normalized, %) | 76.9 |
| Nejumi 4 - HalluLens | 96 | Hallucination resistance (%) | 76 |
| Nejumi 4 - BFCL - Irrelevance Detection | 86.67 | Accuracy (%) | 68.4 |
| Nejumi 4 - GLP - Function Calling | 64.94 | Score (%) | 68 |
| Swallow - English MT-Bench - Writing | 73.7 | Judge Score (normalized, %) | 66.4 |
| Nejumi 4 - jaster (0-shot) - JSICK | 81 | Exact match (%) | 66.3 |
| Swallow - English MT-Bench - Reasoning | 83.5 | Judge Score (normalized, %) | 62.7 |
| Swallow - Post-trained Japanese - PolyMath High and Top | 39.6 | Accuracy (%) | 59.7 |
Interactive version: theaggregate.ai/model?slug=nvidia-nemotron-nano-9b-v2-japanese · How It Works · Data refreshed daily, snapshot 2026-11-05.