Claude Fable 5.2 Leads AutomationBench

Fable 5.2’s agentic-workflow advantage is real, but it is primarily a capability-for-cost tradeoff rather than a clean Opus replacement.

• –AutomationBench evaluates deterministic end states across 47 simulated tools and six business functions, making the result more operationally meaningful than a simple knowledge benchmark.
• –Fable 5.2’s max-effort score reaches 51.4%, while Gemini 4 Carbon remains slightly ahead at 53.9%.
• –Max effort costs $4.64 per task versus $2.29 at xhigh, so the final 3.5-point gain is expensive.
• –The model averages 96.9 tool calls per task at max effort, suggesting its edge comes partly from deeper, more persistent execution.
• –Anthropic still recommends Opus 5.5 as the daily driver, positioning Fable 5.2 for the hardest technical and agentic workloads.