Llama-3-Base-8B-SFT: benchmark results
Princeton NLP's 8B Llama 3 base checkpoint after supervised fine-tuning only, released as the SFT baseline in their SimPO preference-optimization research. Provider: Princeton NLP. Released 2024-05-17. Access: Open.
Aggregate Rating 1357 ± 19, rank #1370 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Portuguese LLM - ENEM | 65.92 | Accuracy (%) | 63 |
| Open Portuguese LLM - OAB Exams | 46.15 | Accuracy (%) | 63 |
| Open Korean LLM - Ko-Winogrande | 61.48 | Accuracy (%) | 62.4 |
| Open Korean LLM - Ko-EQ-Bench | 8.49 | EQ-Bench score | 54.8 |
| Open LLM Leaderboard - GPQA | 29.78 | Score | 53.6 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 21 | Exact match, flexible extract (%) | 53.2 |
| Open Portuguese LLM - BLUEX | 52.29 | Accuracy (%) | 52.2 |
| Open LLM Leaderboard - MuSR | 41.18 | Score | 51.9 |
| Open Korean LLM - KorNAT-Helpful | 46.83 | Normalized accuracy (%) | 49.7 |
| Open Korean LLM Leaderboard | 34.7 | Average Score (%) | 48.5 |
| Open Ko-LLM Leaderboard | 34.7 | Average (%) | 48.4 |
| Open Korean LLM - KorNAT-Harmless | 62.45 | Normalized accuracy (%) | 44.2 |
Interactive version: theaggregate.ai/model?slug=llama-3-base-8b-sft · How It Works · Data refreshed daily, snapshot 2026-10-23.