Llama 3 8B ProLong 64k Base: benchmark results
Princeton NLP's Llama-3-8B ProLong 64K Base, continued-pretrained from Llama 3 8B Instruct for long-context ability, one checkpoint in the ProLong family that scales up to 512K context. Provider: Princeton NLP. Released 2024-07-22. Access: Open.
Aggregate Rating 1432 ± 17, rank #1091 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM Leaderboard | 42.62 | Average Score (%) | 86.7 |
| Open Ko-LLM Leaderboard | 42.62 | Average (%) | 86.6 |
| Open Korean LLM - Ko-GPQA-Diamond | 25.76 | Normalized accuracy (%) | 81.3 |
| Open Korean LLM - Ko-EQ-Bench | 31.09 | EQ-Bench score | 80.2 |
| Open Portuguese LLM - ASSIN2 RTE | 92.44 | Macro F1 (%) | 79 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 42.68 | Exact match, flexible extract (%) | 78 |
| Open Korean LLM - Ko-IFEval | 36.24 | Strict accuracy, prompt/instruction mean (%) | 75.3 |
| Open LLM Leaderboard - MuSR | 43.41 | Score | 73.7 |
| Open Portuguese LLM - OAB Exams | 48.66 | Accuracy (%) | 70.5 |
| Open Portuguese LLM - ENEM | 68.02 | Accuracy (%) | 68.4 |
| Open Portuguese LLM - FaQuAD NLI | 72.26 | Macro F1 (%) | 65.1 |
| Open Korean LLM - KorNAT-Helpful | 48.05 | Normalized accuracy (%) | 63.9 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-prolong-64k-base · How It Works · Data refreshed daily, snapshot 2026-10-20.