Llama 3 70B Instruct: benchmark results
Meta Llama 3 70B instruction-tuned checkpoint. Provider: Meta. Released 2024-04-18. Access: Open.
Aggregate Rating 1543 ± 6, rank #634 of 1656 rated models, from 1157 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| IELTS Writing Score Estimation (Without Examples) - RMSE | 0.99 | RMSE in IELTS band points (0-9), lower is better; predicted | 100 |
| NVIDIA LLM Robustness - AGIEval Consistency | 83.13 | Consistency (%) | 100 |
| NVIDIA LLM Robustness - MMLU-Pro Consistency | 75.24 | Consistency (%) | 100 |
| OpenCompass LLM - Language - English (CompassBench 2405) | 62.9 | Score (%) | 100 |
| OpenCompass LLM - Math - English (CompassBench 2404) | 64.4 | Score (%) | 100 |
| Pak3H - PakAlpaca (Urdu) | 36.2 | Win rate (%; AlpacaEval pairwise preference of the model res | 100 |
| Open LLM Leaderboard v1 - GSM8K | 85.44 | Accuracy (%) (5-shot) | 99.9 |
| Open LLM Leaderboard v1 - MMLU | 80.06 | Accuracy (%) (5-shot) | 99.6 |
| OpenCompass Math - High - English (CompassBench 2404) | 45.2 | Score (%) | 99 |
| Open PL LLM - PSC (generative, 5-shot) | 98.93 | Binary F1 (%) | 98.9 |
| Enkrypt AI - Toxicity Risk | 0 | Risk Score | 98.5 |
| Open CoT - LSAT Logical Reasoning | 22.75 | CoT Gain (%) | 98.5 |
Interactive version: theaggregate.ai/model?slug=llama-3-70b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.