AutomationBench-AA tests AI agents in simulated software workflows
Artificial Analysis says the benchmark scores the share of each task’s objectives completed without guardrail violations.
TLDR
Artificial Analysis describes AutomationBench-AA as a benchmark comparing AI models on tasks across simulated software-as-a-service applications. It measures the share of each task’s objectives completed without guardrail violations, rather than simply whether a task was attempted.
Combined views
709
1 Source, first seen ago