OpenHermes-13B: benchmark results
Teknium's OpenHermes-13B, a fine-tune of Llama 2 13B on the fully open-source, 242,000-entry Hermes dataset combining GPT-4-generated data from GPTeacher, WizardLM, Airoboros and other sources. Provider: Teknium. Released 2023-09-06. Access: Open.
Aggregate Rating 1343 ± 11, rank #1374 of 1635 rated models, from 102 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - College Mathematics | 43 | Accuracy (%) | 99 |
| MMLU-by-task - High School Chemistry | 48.28 | Accuracy (%) | 91.5 |
| MMLU-by-task - High School Statistics | 50 | Accuracy (%) | 90.6 |
| MMLU-by-task - Moral Scenarios | 45.14 | Accuracy (%) | 89.6 |
| MMLU-by-task - International Law | 76.03 | Accuracy (%) | 85.7 |
| MMLU-by-task - Elementary Mathematics | 35.98 | Accuracy (%) | 85.6 |
| MMLU-by-task - Professional Accounting | 45.39 | Accuracy (%) | 85.5 |
| MMLU-by-task - High School Physics | 35.76 | Accuracy (%) | 84.5 |
| MMLU-by-task - Nutrition | 62.75 | Accuracy (%) | 84.1 |
| MMLU-by-task - Moral Disputes | 64.74 | Accuracy (%) | 83.8 |
| MMLU-by-task - High School Microeconomics | 59.24 | Accuracy (%) | 83.7 |
| MMLU-by-task - College Medicine | 53.76 | Accuracy (%) | 83.4 |
Interactive version: theaggregate.ai/model?slug=openhermes-13b · How It Works · Data refreshed daily, snapshot 2026-11-05.