Hermes-3-Llama-3.2-3B: benchmark results
Nous Research's instruction fine-tune of Llama 3.2 3B, using the Hermes 3 training approach. The release card reports training on LambdaLabs H100s but does not disclose enough information to infer training FLOPs. Provider: Nous Research. Released 2024-12-03. Access: Open.
Aggregate Rating 1365 ± 27, rank #7291 of 10916 rated models, from 16 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MuSR | 40.3 | Score | 44.5 |
| Open LLM Leaderboard - IFEval | 38.25 | Score | 37.6 |
| MT-Bench PL - Roleplay | 6.75 | Judge Score (0-10) | 35.7 |
| MT-Bench PL - Coding | 4.45 | Judge Score (0-10) | 32.7 |
| Open LLM Leaderboard - BBH | 43.52 | Score | 30.7 |
| MT-Bench PL - Math | 3.7 | Judge Score (0-10) | 30.6 |
| Open LLM Leaderboard - GPQA | 27.52 | Score | 30.2 |
| MT-Bench PL - STEM | 6.95 | Judge Score (0-10) | 26.5 |
| Open LLM Leaderboard - MMLU-Pro | 25.44 | Score | 26.1 |
| MT-Bench PL - Humanities | 8.05 | Judge Score (0-10) | 24.5 |
| MT-Bench PL - Overall | 5.54 | Judge Score (0-10) | 24.5 |
| Open LLM Leaderboard - MATH Level 5 | 3.93 | Score | 22.5 |
Interactive version: theaggregate.ai/model?slug=hermes-3-llama-3-2-3b · How It Works · Data refreshed daily, snapshot 2026-11-08.