Llama 2 70B: benchmark results
Meta's Llama 2 70B, the base pretrained checkpoint in the Llama 2 series, released under the Llama 2 community license. Provider: Meta. Released 2023-07-18. Access: Open.
Aggregate Rating 1408 ± 10, rank #1201 of 1656 rated models, from 176 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BizBench - TAT-QA Extract | 94.4 | Accuracy (%; 248 TAT-QA questions answerable by extracting a | 100 |
| HELM | 94.35 | Mean win rate (imported) | 100 |
| HELM Classic - NarrativeQA | 76.99 | F1 (%) | 100 |
| HELM Classic - NaturalQuestions Closed Book | 45.84 | F1 (%) | 100 |
| HELM Classic - WikiFact | 44.92 | Exact Match (%) | 100 |
| HELM Classic - bAbI | 70.53 | Exact Match (%) | 100 |
| LLM Benchmarker Suite | 62.53 | Average Score (%) | 100 |
| HELM Classic - BoolQ | 88.6 | Exact Match (%) | 98.5 |
| ToolBench - VirtualHome | 24.74 | Task Score | 97.7 |
| HELM Classic - IMDB | 96.1 | Exact Match (%) | 95.5 |
| HELM Classic - MMLU | 58.17 | Exact Match (%) | 95.5 |
| HELM Classic - QuAC | 48.44 | F1 (%) | 95.4 |
Interactive version: theaggregate.ai/model?slug=llama-2-70b · How It Works · Data refreshed daily, snapshot 2026-10-20.