suzume-llama-3-8B-multilingual: benchmark results
Lightblue's Suzume, a Llama 3 8B Instruct fine-tune on ~90K multilingual conversations to add non-English chat ability to Llama 3. Provider: Lightblue. Released 2024-05-01. Access: Open.
Aggregate Rating 1433 ± 12, rank #1084 of 1656 rated models, from 61 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-IFEval | 48.06 | Strict accuracy, prompt/instruction mean (%) | 92.5 |
| Open Korean LLM - Ko-EQ-Bench | 38.67 | EQ-Bench score | 89.5 |
| Open Korean LLM Leaderboard | 43.77 | Average Score (%) | 88.8 |
| Open Ko-LLM Leaderboard | 43.77 | Average (%) | 88.6 |
| Open LLM Leaderboard v1 - MMLU | 66.62 | Accuracy (%) (5-shot) | 87.7 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 49.13 | Exact match, flexible extract (%) | 82.6 |
| Open Medical LLM - MMLU College Biology | 79.17 | Accuracy (%) | 79.7 |
| Open LLM Leaderboard - IFEval | 66.78 | Score | 78.8 |
| Open Portuguese LLM - ENEM | 70.4 | Accuracy (%) | 76.2 |
| Open Korean LLM - KorNAT-Helpful | 48.93 | Normalized accuracy (%) | 74.9 |
| Open Portuguese LLM - BLUEX | 58.83 | Accuracy (%) | 74.9 |
| Open Medical LLM - MMLU College Medicine | 65.9 | Accuracy (%) | 74.7 |
Interactive version: theaggregate.ai/model?slug=suzume-llama-3-8b-multilingual · How It Works · Data refreshed daily, snapshot 2026-10-20.