Nous-Puffin-70B: benchmark results
Nous Research's Puffin instruction fine-tune of Llama 2 70B, using roughly 3,000 curated examples and a 4,096-token context window. PygmalionAI supplied compute; the card credits Nous Research for the fine-tuning. Provider: Nous Research. Access: Open.
Aggregate Rating 1441 ± 17, rank #4178 of 10916 rated models, from 68 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - Anatomy | 65.93 | Accuracy (%) | 99.6 |
| MMLU-by-task - High School Macroeconomics | 73.33 | Accuracy (%) | 99.6 |
| MMLU-by-task - High School Mathematics | 35.93 | Accuracy (%) | 99.6 |
| MMLU-by-task - Nutrition | 78.1 | Accuracy (%) | 99.6 |
| MMLU-by-task - High School Statistics | 62.96 | Accuracy (%) | 99.5 |
| MMLU-by-task - College Computer Science | 60 | Accuracy (%) | 99 |
| MMLU-by-task - Electrical Engineering | 64.83 | Accuracy (%) | 99 |
| MMLU-by-task - Human Sexuality | 87.02 | Accuracy (%) | 99 |
| MMLU-by-task - Sociology | 89.55 | Accuracy (%) | 99 |
| MMLU-by-task - Medical Genetics | 74 | Accuracy (%) | 98.9 |
| MMLU-by-task - US Foreign Policy | 92 | Accuracy (%) | 98.7 |
| MMLU-by-task - College Biology | 84.03 | Accuracy (%) | 98.6 |
Interactive version: theaggregate.ai/model?slug=nous-puffin-70b · How It Works · Data refreshed daily, snapshot 2026-11-08.