Llama 2 7B 32K: benchmark results
Together's long-context fine-tune of Meta's Llama 2 7B, extended to a 32K-token context via position interpolation and FlashAttention-2, for multi-document QA and long-text summarization. Provider: Together. Released 2023-07-18. Access: Open.
Aggregate Rating 1264 ± 12, rank #1513 of 1640 rated models, from 83 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MMLU-by-task - College Mathematics | 34 | Accuracy (%) | 69.2 |
| MMLU-by-task - College Computer Science | 44 | Accuracy (%) | 67.6 |
| MMLU-by-task - Global Facts | 35 | Accuracy (%) | 64.2 |
| MMLU-by-task - College Chemistry | 37 | Accuracy (%) | 62.4 |
| MMLU-by-task - Elementary Mathematics | 30.42 | Accuracy (%) | 59.1 |
| MMLU-by-task - Medical Genetics | 54 | Accuracy (%) | 58.3 |
| MMLU-by-task - Econometrics | 28.95 | Accuracy (%) | 55.2 |
| MMLU-by-task - High School Statistics | 39.81 | Accuracy (%) | 54.3 |
| MMLU-by-task - Astronomy | 44.74 | Accuracy (%) | 52.6 |
| MMLU-by-task - High School US History | 59.31 | Accuracy (%) | 51.8 |
| MMLU-by-task - High School World History | 64.56 | Accuracy (%) | 51.4 |
| MMLU-by-task - High School Physics | 30.46 | Accuracy (%) | 50.8 |
Interactive version: theaggregate.ai/model?slug=llama-2-7b-32k · How It Works · Data refreshed daily, snapshot 2026-10-13.