Newest papers

Bridging Foundational and Agentic Capabilities of LLMs

Wei Guo, Yuanjian Xu, Jianing Hao, Yongyao Duan, Ran Chen, Zhong Li, Guang Zhang · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29751 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Los grandes modelos de lenguaje (LLM) se despliegan cada vez más como agentes, pero sigue siendo poco claro cómo sus capacidades fundamentales se traducen en capacidades de agentes. Las puntuaciones de fin a fin dejan sin resolver si la capacidad de referencia baja o la implantación débil conduce a un fracaso, limitando el diagnóstico actuable. Brindamos las capacidades fundacionales y de agentes a través de un marco de medición a nivel de operación. Representamos cada modelo con un perfil fundacional derivado de los parámetros de la capacidad y un perfil de operación derivado de las trayectorias de interacción, y definimos las brechas de realización de la capacidad como desviaciones del rendimiento observado de las predicciones basadas en las capacidades fundamentales. Los experimentos en el modelo de servicios de $^2$-

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

Abstract

Large language models (LLMs) are increasingly deployed as agents, yet it remains unclear how their foundational capabilities translate into agentic capabilities. End-to-end scores leave unresolved whether low benchmark capability or weak deployment drives a failure, limiting actionable diagnosis. We bridge foundational and agentic capabilities through an operation-level measurement framework. We represent each model with a foundational profile derived from capability benchmarks and an operation profile derived from interaction trajectories, and define capability realization gaps as deviations of observed operation performance from predictions based on foundational capabilities. Experiments on fifteen model services in $τ^2$-Bench show that foundational profiles only partially predict agentic operations. Models with comparable profiles can differ substantially in execution, and these gaps remain consistent across task subsets. Operation profiles add information and substantially improve task-score prediction. The framework locates execution departures at the operation level to inform model development and agent evaluation. We will release our code and analysis artifacts.

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

Topics: Bridging (networking), Limiting, Computer science, Bridge (graph theory), Software deployment, Task (project management)

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