Llama 3 8B ProLong 64k Instruct: benchmark results
Princeton NLP's ProLong 64K, Llama 3 8B Instruct continued-trained on 20B tokens of long-context data and SFT'd on UltraChat. Provider: Princeton NLP. Released 2024-07-21. Access: Open.
Aggregate Rating 1453 ± 17, rank #981 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM Leaderboard | 44.61 | Average Score (%) | 90.7 |
| Open Ko-LLM Leaderboard | 44.62 | Average (%) | 90.5 |
| Open Korean LLM - KorNAT-Helpful | 51.33 | Normalized accuracy (%) | 90.1 |
| Open Korean LLM - Ko-GPQA-Diamond | 26.77 | Normalized accuracy (%) | 88.9 |
| Open Korean LLM - Ko-EQ-Bench | 36.3 | EQ-Bench score | 86.6 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 49.58 | Exact match, flexible extract (%) | 83.3 |
| Open Korean LLM - Ko-IFEval | 38.75 | Strict accuracy, prompt/instruction mean (%) | 80.2 |
| Open LLM Leaderboard - MuSR | 43.97 | Score | 77.8 |
| Open Korean LLM - KorNAT-Harmless | 64.5 | Normalized accuracy (%) | 74.6 |
| Open Portuguese LLM - ASSIN2 RTE | 92.07 | Macro F1 (%) | 73.9 |
| Open Portuguese LLM - ENEM | 68.3 | Accuracy (%) | 69.8 |
| Open LLM Leaderboard - IFEval | 55.63 | Score | 67.4 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-prolong-64k-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.