WorkflowArena: Benchmarking Executable Workflow Orchestration in LLM-Based Multi-Agent Systems
Workflows authored by Claude Opus 5.5 substantially improved weaker executors on a challenging subset of tasks.
WorkflowArena tests whether models can organize delegated work across 145 long-horizon tasks in eight professional domains. Agents choose whether and how to delegate, while hidden verifiers assess the integrated deliverables and delegation records capture how the work was organized. Experiments with nine models found that no single configuration performed best across all domains. Holding the executing model fixed helped isolate workflow organization as a measurable contributor to task performance.
ArXiv · AI/CL/LG's note
WorkflowArena tests whether models can organize delegated work across 145 long-horizon tasks in eight professional domains. Agents choose whether and how to delegate, while hidden verifiers assess the integrated deliverables and delegation records capture how the work was organized. Experiments with nine models found that no single configuration performed best across all domains. Holding the executing model fixed helped isolate workflow organization as a measurable contributor to task performance.
ArXiv · AI/CL/LG's note
score 5