Newest papers

StepBench: Diagnosing Real-World Procedural Understanding in AI Agents

Hossein A. Rahmani, Bin Wu, Chaeeun Lee, Fanghua Ye, Sahan Bulathwela, Sujay Kumar Jauhar, Marzieh Fadaee, Nick Craswell, Emine Yılmaz · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.26679 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Evaluar la capacidad de planificación de un agente sigue siendo un desafío para las tareas de dominio abierto, donde la verificación de los procedimientos a través de la ejecución a menudo es poco práctica. Las evaluaciones existentes suelen evaluar solo el plan final y no distinguir errores procedimentales tales como falta, incorrecto o mal ordenado de las acciones. Introducimos STEPBENCH, un benchmark de 6.885 procedimientos verificados recorridos de wikiHow a través de 81 categorías de dominio abierto. De cada procedimiento, derivamos 5 tareas de diagnóstico: cuatro con respuestas deterministas que involucran medidas retenidas, incorrectas o mal ordenadas, y uno que requiere la próxima generación de pasos. Además evaluamos la generación de plan completo para examinar cómo el rendimiento diagnóstico se correla

Machine translation (Meta M2M100); the English original follows.

Abstract

Evaluating an agent's planning ability remains challenging for open-domain tasks, where verifying procedures through execution is often impractical. Existing evaluations typically assess only the final plan and do not distinguish procedural errors such as missing, incorrect, or misordered actions. We introduce STEPBENCH, a benchmark of 6,885 verified procedures crawled from wikiHow across 81 open-domain categories. From each procedure, we derive 5 diagnostic tasks: four with deterministic answers involving withheld, incorrect, or shuffled steps, and one requiring next-step generation. We additionally evaluate full-plan generation to examine how diagnostic performance correlates with end-to-end plan validity. Using the same procedures across tasks lets us directly compare different planning capabilities. For generation, STEPEVAL identifies critical failures that prevent goal completion. Across 14 configurations of 12 models, we find substantial differences across planning capabilities, with step ordering proving the most difficult: the strongest model correctly orders only 63.0% of procedures, dropping to 33.1% for those with nine or more steps. Incorrect-step detection also predicts judged plan validity across models, suggesting a lower-cost alternative for evaluating generation quality. On STEPBENCH-Val, comprising 214 procedures published in 2026, comprehension scores and rankings remain stable, suggesting limited influence from memorisation.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Plan (archaeology), Artificial intelligence, Machine learning, Benchmark (surveying), Comprehension

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange