Llama-3-Instruct-8B-SimPO: benchmark results
Princeton NLP's Llama 3 8B Instruct preference-tuned with reference-free SimPO - the main checkpoint from the SimPO paper. Provider: Princeton NLP. Released 2024-05-17. Access: Open.
Aggregate Rating 1440 ± 14, rank #1048 of 1656 rated models, from 33 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - IFEval | 73.47 | Score | 88 |
| Open Korean LLM - KorNAT-Helpful | 50.72 | Normalized accuracy (%) | 86.8 |
| Open Korean LLM - Ko-GPQA-Diamond | 26.26 | Normalized accuracy (%) | 85.5 |
| BenchBench | 79.92 | Aggregate Score (%) | 84.6 |
| AlpacaEval 2.0 | 44.65 | LC Win Rate (%) | 84.2 |
| Open Korean LLM - Ko-IFEval | 40.69 | Strict accuracy, prompt/instruction mean (%) | 83.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 46.32 | Exact match, flexible extract (%) | 80.6 |
| Open Korean LLM - KorNAT-Harmless | 64.61 | Normalized accuracy (%) | 75.4 |
| Open Korean LLM - KorNAT-CKA | 27.75 | Normalized accuracy (%) | 71.4 |
| Open Portuguese LLM - OAB Exams | 48.47 | Accuracy (%) | 69.8 |
| Open Portuguese LLM - BLUEX | 57.3 | Accuracy (%) | 69.4 |
| Open Chinese LLM - GSM8K | 49.43 | Accuracy (%) | 65.1 |
Interactive version: theaggregate.ai/model?slug=llama-3-instruct-8b-simpo · How It Works · Data refreshed daily, snapshot 2026-10-20.