ConfigurableHermes-7B: benchmark results
vicgalle's ConfigurableHermes 7B, a Hermes-based fine-tune using configurable safety tuning so system prompts can switch between harmless, uncensored and persona behaviors. Provider: vicgalle. Released 2024-02-17. Access: Open.
Aggregate Rating 1393 ± 15, rank #1250 of 1656 rated models, from 35 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - KorNAT-Harmless | 67.6 | Normalized accuracy (%) | 90.1 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 61.71 | MC2 (%) (0-shot) | 78.1 |
| Open Chinese LLM - TruthfulQA MC | 58.47 | Accuracy (%) | 76.6 |
| Open LLM Leaderboard v1 - GSM8K | 61.41 | Accuracy (%) (5-shot) | 74 |
| Open LLM Leaderboard v1 - ARC Challenge | 66.04 | Normalized accuracy (%) (25-shot) | 72.1 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 34.72 | Exact match, flexible extract (%) | 67.8 |
| Open LLM Leaderboard v1 - HellaSwag | 84.31 | Normalized accuracy (%) (10-shot) | 65.3 |
| Open LLM Leaderboard - IFEval | 54.11 | Score | 64.9 |
| Open Korean LLM - Ko-IFEval | 32.2 | Strict accuracy, prompt/instruction mean (%) | 61.7 |
| Open Portuguese LLM - FaQuAD NLI | 70.41 | Macro F1 (%) | 61.1 |
| Open Korean LLM Leaderboard | 36.88 | Average Score (%) | 59.8 |
| Open Ko-LLM Leaderboard | 36.88 | Average (%) | 59.7 |
Interactive version: theaggregate.ai/model?slug=configurablehermes-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.