HarmActionsEval: leaderboard
Agent safety benchmark measuring how often autonomous LLM agents avoid harmful tool actions under adversarial pressure.
Metric: SafeActions@1 (self-reported). Source: benchmarklist.com. Status: saturation imminent. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 397B A17B | 23.4 |
| 2 | GPT-5.3 | 12.77 |
| 3 | Claude Sonnet 4.6 | 2.84 |
| 4 | Phi-4-mini (Reasoning) | 2.84 |
| 5 | Ministral 3 3B | 2.13 |
| 6 | Gemini 3.1 Flash Lite | 0.71 |
| 7 | GPT-5.4 Mini | 0.71 |
| 8 | Claude Haiku 4.5 | 0 |
| 9 | Phi-4 Mini Instruct | 0 |
Interactive version: theaggregate.ai/benchmark?slug=harmactionseval · How It Works · Data refreshed daily, snapshot 2026-09-05.