janus-orpo-7B: benchmark results
KAIST AI's 7B model, created by applying ORPO to Mistral-7B-v0.2 with a preference dataset of 196,000 unique system messages for personalized-preference alignment. Provider: KAIST. Released 2024-04-18. Access: Open.
Aggregate Rating 1366 ± 17, rank #1348 of 1656 rated models, from 23 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-EQ-Bench | 23.98 | EQ-Bench score | 69.3 |
| Open Korean LLM - KorNAT-Helpful | 48.25 | Normalized accuracy (%) | 66.3 |
| Open Portuguese LLM - FaQuAD NLI | 71.02 | Macro F1 (%) | 62.6 |
| Open Korean LLM - Ko-IFEval | 30.62 | Strict accuracy, prompt/instruction mean (%) | 57.3 |
| Open Korean LLM - KorNAT-Harmless | 63.13 | Normalized accuracy (%) | 56.9 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 22.67 | Exact match, flexible extract (%) | 55.2 |
| Open Korean LLM Leaderboard | 35.61 | Average Score (%) | 52.5 |
| Open Ko-LLM Leaderboard | 35.61 | Average (%) | 52.4 |
| Open Portuguese LLM - ASSIN2 RTE | 89.3 | Macro F1 (%) | 49.6 |
| Open Portuguese LLM - BLUEX | 49.37 | Accuracy (%) | 45 |
| Open Portuguese LLM - OAB Exams | 38.91 | Accuracy (%) | 39.5 |
| Open Portuguese LLM - ENEM | 56.12 | Accuracy (%) | 39.3 |
Interactive version: theaggregate.ai/model?slug=janus-orpo-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.