MedBench v5 - SafetyAgent: leaderboard

Metric: Environment score (0-100) on SafetyAgent (red-team robustness in a sandbox built on OpenRT: medical misinformation, dangerous tool commands, privilege escalation, privacy leakage and ethics violations; interception ratio and agent-as-a-judge); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.

Top models

#ModelScore
1GPT-5.596
2Kimi K2.680
3Seed 2.0 Pro60.67

Interactive version: theaggregate.ai/benchmark?slug=medbench-v5-safetyagent · How It Works · Data refreshed daily, snapshot 2026-09-29.