SREGym (No Noise): leaderboard
SREGym (Clark et al., 2026) injects 90 failure scenarios into live Kubernetes deployments, from application misconfigurations to operating-system, hardware, metastable and concurrent failures, and an AI agent must diagnose and mitigate them through metrics, logs, traces and cluster commands. End-to-end success rate in percent over three runs per problem: a run counts only when both the diagnosis and the mitigation succeed, higher is better. No ambient noise is injected.
Metric: End-to-end success rate (%; runs with both a correct diagnosis and a successful mitigation, 90 problems x 3 runs, without injected noise). Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 60.7 | #158 |
| 2 | GPT-5.4 (2026-03-05) | 53.3 | #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-no-noise · How It Works · Data refreshed daily, snapshot 2026-11-12.