Llama 2 7B: benchmark results
Meta's Llama 2 7B, the base pretrained checkpoint released under the Llama 2 Community License in July 2023; the linked model card provides no further details here. Provider: Meta. Released 2023-07-18. Access: Open.
Aggregate Rating 1269 ± 10, rank #1520 of 1656 rated models, from 196 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SAD - Predict Words | 61.22 | Self-prediction score (%, higher is better) | 100 |
| SAD - Predict Words (Situating Prompt) | 57.89 | Self-prediction score (%, higher is better) | 100 |
| SAD - Introspection | 37.95 | Score (%, higher is better) | 90 |
| SAD - Introspection (Situating Prompt) | 37.32 | Score (%, higher is better) | 90 |
| HELM Classic - Synthetic Reasoning Natural | 26.02 | F1 (%) | 82.4 |
| HELM Classic - bAbI | 54.97 | Exact Match (%) | 81.9 |
| HELM Classic - QuAC | 40.62 | F1 (%) | 80 |
| ToolBench - WebShop Short | 6.92 | Task Score | 79.1 |
| HELM Classic - WikiFact | 33.49 | Exact Match (%) | 75.8 |
| HELM Classic - Dyck | 65.8 | Exact Match (%) | 75 |
| HELM Classic - Entity Matching | 85.07 | Exact Match (%) | 74.2 |
| HELM Classic - MATH Chain-of-Thought | 9.91 | Equivalent (%) | 72.1 |
Interactive version: theaggregate.ai/model?slug=llama-2-7b · How It Works · Data refreshed daily, snapshot 2026-10-20.