Vector Eval - AgentHarm-Benign: leaderboard
Vector Institute Inspect Evals run of AgentHarm-Benign, checking whether agents can still complete harmless tasks while maintaining safety behavior.
Metric: Average Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 82.49 | #343 |
| 2 | Claude 3.5 Sonnet | 80.07 | #350 |
| 3 | O1 | 72.35 | #183 |
| 4 | Gemini 1.5 Pro | 59.61 | #445 |
| 5 | O3 Mini | 54.29 | #272 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vector-eval-agentharm-benign · How It Works · Data refreshed daily, snapshot 2026-10-20.