Nous-Hermes-13B: benchmark results
Nous Research's Nous-Hermes-13B, a Llama 13B fine-tune on over 300,000 mostly GPT-4-generated instructions from GPTeacher, CodeAlpaca, Airoboros and other sources, trained by Teknium and Karan4D. Provider: Nous Research. Released 2023-06-03. Access: Open.
Aggregate Rating 1305 ± 11, rank #1454 of 1640 rated models, from 77 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - Global Facts | 40 | Accuracy (%) | 88.7 |
| MMLU-by-task - Formal Logic | 38.89 | Accuracy (%) | 83.9 |
| MMLU-by-task - TruthfulQA MC2 | 51.5 | Accuracy (%) | 83.2 |
| MMLU-by-task - TruthfulQA MC1 | 35.37 | Accuracy (%) | 82.6 |
| MMLU-by-task - Human Sexuality | 63.36 | Accuracy (%) | 76.4 |
| MMLU-by-task - HellaSwag | 62.22 | Accuracy (%) | 75.1 |
| MMLU-by-task - Machine Learning | 37.5 | Accuracy (%) | 71.9 |
| MMLU-by-task - Management | 71.84 | Accuracy (%) | 71.7 |
| MMLU-by-task - US Foreign Policy | 79 | Accuracy (%) | 70 |
| MMLU-by-task - ARC Challenge | 54.95 | Accuracy (%) | 69 |
| MMLU-by-task - Sociology | 71.14 | Accuracy (%) | 69 |
| MMLU-by-task - Professional Psychology | 52.78 | Accuracy (%) | 67.6 |
Interactive version: theaggregate.ai/model?slug=nous-hermes-13b · How It Works · Data refreshed daily, snapshot 2026-10-13.