Llama 3.1 70B: benchmark results
Meta's open Llama 3.1 70B dense model (July 2024). Provider: Meta. Released 2024-07-23. Access: Open.
Aggregate Rating 1557 ± 7, rank #592 of 1656 rated models, from 669 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AbstentionBench - stale - FreshQA - Precision | 62.5 | Abstention Precision (%) | 100 |
| AbstentionBench - subjective - MoralChoice - F1 | 56.73 | Abstention F1 (%) | 100 |
| AbstentionBench - subjective - MoralChoice - Recall | 57.35 | Abstention Recall (%) | 100 |
| AbstentionBench - underspecified context - BB/Disambiguate - F1 | 66.99 | Abstention F1 (%) | 100 |
| AbstentionBench - underspecified context - Musique - Precision | 99.41 | Abstention Precision (%) | 100 |
| AbstentionBench - underspecified context - SQuAD 2.0 - Precision | 99.39 | Abstention Precision (%) | 100 |
| EuroEval German NLU | 65.97 | NLU Average Score (%) | 100 |
| EuroEval Greek NLU - MultiWikiQA EL | 76.53 | Character F1 (%) | 100 |
| EuroEval Greek Summarization - Greek Wikipedia | 36.77 | chrF3++ (%) | 100 |
| EuroEval Hungarian NLU | 65.84 | NLU Average Score (%) | 100 |
| Polar (South Korea) - Sociocultural Axis | 96.23 | Sociocultural-axis ICAT (0-100), the StereoSet idealized con | 100 |
| SeaEval - Cross-Lingual Consistency - Cross-LogiQA (Few-Shot) | 62.42 | Accuracy (%) | 100 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-70b · How It Works · Data refreshed daily, snapshot 2026-10-20.