InfraBench evaluates AI agents on realistic infrastructure incidents across the full stack (L1 hardware → L4 applications) and the operational lifecycle — durability, invariants, cleanup, and risk — not just binary pass/fail.
Website / leaderboard:
https://infraben.ch
Paper (PDF): Beyond Pass/Fail: Evaluating Infrastructure Agents Across Layers, Lifecycle, and Risk · HotInfra '26 (co-located with ISCA '26)
GitHub: XuanmiaoG/InfraBench
