suzume-llama-3-8B-multilingual-orpo-borda-top25: benchmark results
Lightblue's ORPO-tuned Suzume Llama 3 8B, preference-trained on the top 25% most consistently ranked responses from its Mitsu dataset. Provider: Lightblue. Released 2024-04-26. Access: Open.
Aggregate Rating 1459 ± 17, rank #944 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 53.33 | Strict accuracy, prompt/instruction mean (%) | 95.9 |
| Open Korean LLM - Ko-EQ-Bench | 39.48 | EQ-Bench score | 90.2 |
| Open Ko-LLM Leaderboard | 43.87 | Average (%) | 89.3 |
| Open Korean LLM Leaderboard | 43.64 | Average Score (%) | 88.5 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 49.2 | Exact match, flexible extract (%) | 82.8 |
| Open Korean LLM - KorNAT-Harmless | 65.82 | Normalized accuracy (%) | 82.7 |
| Open Portuguese LLM - ENEM | 71.03 | Accuracy (%) | 78.8 |
| Open LLM Leaderboard - IFEval | 66.37 | Score | 78.4 |
| Open Portuguese LLM - BLUEX | 60.08 | Accuracy (%) | 77.4 |
| Open Portuguese LLM - OAB Exams | 49.75 | Accuracy (%) | 74.6 |
| Open Korean LLM - KorNAT-CKA | 27.96 | Normalized accuracy (%) | 72.9 |
| Open Portuguese LLM - ASSIN2 RTE | 91.87 | Macro F1 (%) | 71.4 |
Interactive version: theaggregate.ai/model?slug=suzume-llama-3-8b-multilingual-orpo-borda-top25 · How It Works · Data refreshed daily, snapshot 2026-10-20.