Enhancing Agent Safety Judgment: leaderboard
Tool-using agent systems powered by large language models (LLMs) are increasingly deployed across web, app, operating-system, and transactional environments. Yet existing safety benchmarks still emphasize explicit risks, potentially overstating a model's ability to judge deceptive or ambiguous trajectories. To address this gap, we introduce ROME (Red-team Orchestrated Multi-agent Evolution), a con
Metric: Original (Seed) (self-reported). Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 | 94.53 | #251 |
| 2 | GPT-4o | 93.68 | #343 |
| 3 | Claude 3.7 Sonnet | 90.59 | #249 |
| 4 | Qwen 3 235B A22B | 89.42 | #313 |
| 5 | DeepSeek V3 | 88.24 | #321 |
| 6 | Qwen 3 8B | 81.82 | #674 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=enhancing-agent-safety-judgment · How It Works · Data refreshed daily, snapshot 2026-10-20.