Llama 3 8B ProLong 512k Base: benchmark results
Princeton NLP's Llama-3-8B ProLong 512K Base, continued-trained from their own ProLong 64K Base checkpoint to extend context further, in the family's 512K-max-context line. Provider: Princeton NLP. Released 2024-08-22. Access: Open.
Aggregate Rating 1427 ± 17, rank #1116 of 1656 rated models, from 24 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Ko-LLM Leaderboard | 42.29 | Average (%) | 85.6 |
| Open Korean LLM Leaderboard | 42.29 | Average Score (%) | 85.6 |
| Open Korean LLM - Ko-GPQA-Diamond | 25.76 | Normalized accuracy (%) | 81.3 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 42.84 | Exact match, flexible extract (%) | 78.3 |
| Open Korean LLM - Ko-IFEval | 36.82 | Strict accuracy, prompt/instruction mean (%) | 76.7 |
| Open Korean LLM - Ko-EQ-Bench | 28.18 | EQ-Bench score | 75.6 |
| Open LLM Leaderboard - IFEval | 53.22 | Score | 63.3 |
| Open LLM Leaderboard - MuSR | 42.23 | Score | 62.7 |
| Open Korean LLM - Ko-Winogrande | 61.33 | Accuracy (%) | 61.4 |
| Open Korean LLM - KorNAT-Harmless | 63.13 | Normalized accuracy (%) | 56.7 |
| Open Korean LLM - KorNAT-Helpful | 47.4 | Normalized accuracy (%) | 56.3 |
| Open LLM Leaderboard - BBH | 50.33 | Score | 49.6 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-prolong-512k-base · How It Works · Data refreshed daily, snapshot 2026-10-20.