AceReason-Nemotron-14B: benchmark results
NVIDIA's math and code reasoning derivative of DeepSeek-R1-Distill-Qwen-14B, post-trained with reinforcement learning on math prompts followed by code prompts. It is distinct from the later AceReason-Nemotron-1.1 releases. Provider: NVIDIA. Access: Open.
Aggregate Rating 1501 ± 35, rank #2350 of 10916 rated models, from 50 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Japanese LLM - Xlsum JA Rouge1 | 30.32 | Score (%) | 83.2 |
| Open Japanese LLM - Jsick Exact Match | 83.19 | Score (%) | 80.6 |
| Open Japanese LLM - EL | 57.32 | Score (%) | 75.1 |
| Open Japanese LLM - SUM | 10.61 | Score (%) | 72.8 |
| Open Japanese LLM - Xlsum JA Rouge2 | 10.61 | Score (%) | 72.7 |
| Open Japanese LLM - Wiki NER SET F1 | 8.85 | Score (%) | 72 |
| Open Japanese LLM - Xlsum JA Bert Score JA F1 | 69.67 | Score (%) | 66.7 |
| Open Japanese LLM - Jnli Exact Match | 82.25 | Score (%) | 56.9 |
| Open Japanese LLM - Jsem Exact Match | 77.53 | Score (%) | 56.4 |
| Open Japanese LLM - NLI | 76.19 | Score (%) | 56.1 |
| Open Japanese LLM - Wiki Coreference SET F1 | 3.09 | Score (%) | 53.1 |
| Open Japanese LLM - CG | 13.45 | Score (%) | 52.4 |
Interactive version: theaggregate.ai/model?slug=acereason-nemotron-14b · How It Works · Data refreshed daily, snapshot 2026-11-08.