Llama 3 8B Instruct: benchmark results
Meta Llama 3 8B instruction-tuned checkpoint. Provider: Meta. Released 2024-04-18. Access: Open.
Unified ELO 1412 ± 1, rank #1091 of 1392 rated models, from 472 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CyberSecEval | 45.27 | Average Injection Success Rate (self-reported) | 100 |
| HumanLikeness - Syntax-2 | 79.49 | Humanlike Score (%) | 100 |
| Open CoT - LSAT Reading Comprehension | 25.65 | CoT Gain (%) | 99.2 |
| AGC-Bench - mops | 2.05 | Dataset z-score | 98.8 |
| LLM Trustworthy - Adversarial | 67.28 | Trust Score (%) | 98 |
| LLM Trustworthy - Ethics | 93.74 | Trust Score (%) | 98 |
| Open CoT Leaderboard | 14.75 | Average CoT Gain (%) | 95.4 |
| AGC-Bench - creatset | 1.48 | Dataset z-score | 95.1 |
| Open LLM Leaderboard - MuSR | 19.94 | Score | 94.7 |
| Open CoT - LogiQA | 8.15 | CoT Gain (%) | 90.5 |
| LLM Trustworthy Leaderboard | 80.61 | Average Trust Score (%) | 90 |
| EuroEval Portuguese NLU - HAREM | 56.38 | Named entity recognition Score (%) | 89.9 |
Interactive version: theaggregate.ai/model?slug=llama-3-8b-instruct · How It Works · Data refreshed daily, snapshot 2026-09-05.