Llama 2 13B Chat Base: benchmark results
Meta's Llama 2 13B Chat, the RLHF fine-tuned chat version of Llama 2 13B, released under the Llama 2 Community License in July 2023; the linked card provides no further details here. Provider: Meta. Released 2023-07-18. Access: Open.
Aggregate Rating 1333 ± 9, rank #1422 of 1656 rated models, from 236 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HumanLikeness - Discourse-2 | 78.85 | Humanlike Score (%) | 100 |
| HumanLikeness - Word-2 | 25.05 | Humanlike Score (%) | 94.7 |
| SALAD-Bench | 81.27 | Average Safety Score (%) | 93.9 |
| AdvBench | 1 | AdvBench harmful-instruction score | 92.9 |
| Enkrypt AI - Toxicity Risk | 0.45 | Risk Score | 90.8 |
| SALAD-Bench Base | 96.81 | Safety Score (%) | 89.4 |
| MMLU-by-task - High School Chemistry | 46.31 | Accuracy (%) | 88.6 |
| SALAD-Bench Attack | 65.72 | Safety Score (%) | 87.9 |
| MMLU-by-task - International Law | 76.86 | Accuracy (%) | 87.2 |
| Enkrypt AI - Jailbreak Risk | 3.29 | Risk Score | 86.1 |
| MMLU-by-task - Public Relations | 66.36 | Accuracy (%) | 85.5 |
| HumanLikeness - Meaning-1 | 73.9 | Humanlike Score (%) | 84.2 |
Interactive version: theaggregate.ai/model?slug=llama-2-13b-chat-base · How It Works · Data refreshed daily, snapshot 2026-10-20.