SREGym (Noise Injected) - Mitigation: leaderboard
Mitigation success on the 90 SREGym failure scenarios (Clark et al., 2026): a mitigation oracle checks the live system after the agent acts, in percent of runs (three per problem), higher is better. Two transient ambient noise events, such as a crashing pod or dropped requests, are injected every five minutes alongside the target fault.
Metric: Mitigation success rate (%; mitigation oracle on the live system, 90 problems x 3 runs, with two ambient noise patterns injected every five minutes). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 76.3 | #158 |
| 2 | GPT-5.4 (2026-03-05) | 61.9 | #84 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=sregym-noise-injected-mitigation · How It Works · Data refreshed daily, snapshot 2026-11-12.