CloudOpsBench: AI benchmark shows FAI models struggling with cloud infra

The CloudOpsBench v1.0 leaderboard shows the top AI agent configuration, claude-fable-5-1 (max) run through claude-code, resolving only 60.6% of real cloud operations tasks, at a cost of $404.3 and 38.6 hours across 97.9M tokens. The next closest entries were gpt-6-astra (max) via codex at 58.9% resolution ($310.9, 16.5h, 98.0M tokens) and gemini-4-argon (high) via gemini-cli at 57.5% ($100.5, 34.2h, 173.9M tokens). The results indicate that even the strongest agent harnesses fail roughly four in ten cloud infrastructure tasks, with grok-4.7 (high) on cursor-cli at 56.5% and claude-opus-5-5 (high) on terminus-2 at 47.8% rounding out the published rankings.