AgentS4D (Hermes Agent): leaderboard
AgentS4D (Zhejiang University of Technology, 2026) runtime-safety cases for workspace agents: 328 risk-injected cases from 76 Workspace-Bench tasks, run in the Hermes agent harness. The score is the conditional attack success rate in percent, unsafe runs over runs with an unsafe signal, explicit defense or confirmed payload contact; lower is better.
Metric: Conditional attack success rate (%, cASR): unsafe runs over runs that were unsafe, explicitly defended, or safe after confirmed payload contact, over the 328 risk-injected cases run once in the Hermes agent (0.14.0) harness; lower is better. Source: arxiv.org. Saturation forecast: Around 2029. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.7 Plus | 58.13 | #207 |
| 2 | MiniMax-M3 | 64.69 | #269 |
| 3 | GPT-5.5 | 70.67 | #49 |
| 4 | Gemini 3.1 Pro (Preview) | 90.28 | #94 |
| 5 | DeepSeek V4 Pro | 91.02 | #188 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=agents4d-hermes-agent · How It Works · Data refreshed daily, snapshot 2026-11-13.