Llama 3.1 Nemotron 70B Instruct: benchmark results
NVIDIA's RLHF-tuned Llama 3.1 70B (HelpSteer2-Preference + REINFORCE), topping Arena Hard and AlpacaEval 2 at release (October 2024). Provider: NVIDIA. Released 2024-10-15. Access: Open.
Aggregate Rating 1567 ± 11, rank #554 of 1656 rated models, from 321 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open PL LLM - DYK (generative, 5-shot) | 71.82 | Binary F1 (%) | 98.1 |
| Open PL LLM - Belebele (multiple choice, 5-shot) | 92.44 | Accuracy (%) | 97.8 |
| EuroEval Portuguese | 68.25 | Average Score (%) | 97.7 |
| Open Japanese LLM - MMLU EN Exact Match | 81.23 | Score (%) | 97.6 |
| Open Japanese LLM - Jsick Exact Match | 86.62 | Score (%) | 97.1 |
| Open PL LLM - 8TAGS (generative, 5-shot) | 79.78 | Accuracy (%) | 97 |
| Open CoT - LSAT Logical Reasoning | 21.96 | CoT Gain (%) | 96.9 |
| Open PL LLM - Belebele (generative, 5-shot) | 92 | Accuracy (%) | 96.9 |
| Open PL LLM - Generative | 69.13 | Average Generative Score (%) | 96.9 |
| StickToYourRole | 80.66 | Cardinal Score | 96.8 |
| Stick To Your Role! | 0.81 | Cardinal Score (-1 to 1) | 96.7 |
| Open PL LLM - 8TAGS (multiple choice, 5-shot) | 80.01 | Accuracy (%) | 96.2 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-nemotron-70b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.