Llama 3.1 8B: benchmark results
Meta's 8B pretrained base model from the Llama 3.1 family (July 2024), with 128K context and trained on over 15T tokens. Provider: Meta. Released 2024-07-23. Access: Open.
Aggregate Rating 1420 ± 5, rank #1148 of 1656 rated models, from 1251 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AbstentionBench - stale - FreshQA - F1 | 52.08 | Abstention F1 (%) | 100 |
| CHARM - Boundary Awareness | 94.1 | Accuracy (%) | 100 |
| SeaEval - Fundamental NLP Tasks - C3 (Few-Shot) | 81.04 | Accuracy (%) | 100 |
| When Context Flips, Safety Breaks | 77.6 | PacifAIst BSR (self-reported) | 100 |
| LA Leaderboard - CoQCat | 76.67 | F1 (%) | 98.5 |
| EuroEval Swedish NLU - Swerec | 80.44 | Matthews correlation coefficient (%) | 97.5 |
| LA Leaderboard - RagQuAS | 81.39 | Semantic Answer Similarity (x100) | 97.1 |
| EuroEval German Summarization - Mlsum DE | 38.51 | chrF3++ (%) | 95.9 |
| EuroEval Polish NLU - PoQuAD | 61.78 | Character F1 (%) | 95.7 |
| EuroEval Lithuanian NLU - MultiWikiQA LT | 72.09 | Character F1 (%) | 95.6 |
| LA Leaderboard - XQuAD Catalan | 74.64 | F1 (%) | 95.4 |
| EuroEval Greek NLU - MultiWikiQA EL | 72.93 | Character F1 (%) | 95.1 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-8b · How It Works · Data refreshed daily, snapshot 2026-10-21.