SafeArena: leaderboard
Safety benchmark for autonomous web agents over safe and harmful web tasks, including normalized safety, harmful completion, safe completion, and refusal rates.
Metric: Normalized Safety Score (imported). Source: arxiv.org. Source: benchmarklist.com. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet (20240620) | 55 | #299 |
| 2 | GPT-4o Mini | 35.7 | #599 |
| 3 | Llama 3.2 90B Vision Instruct | 34 | #622 |
| 4 | GPT-4o | 31.7 | #343 |
| 5 | Qwen 2 VL 72B | 21.5 | #460 |
Interactive version: theaggregate.ai/benchmark?slug=safearena · How It Works · Data refreshed daily, snapshot 2026-10-20.