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
Les grands modèles linguistiques (LLM) sont de plus en plus déployés en tant qu'agents, mais il reste peu clair comment leurs capacités fondamentales se traduisent en capacités d'agent. Les scores de bout en bout ne sont pas résolus si la capacité de référence faible ou le déploiement faible entraîne un échec, limitant le diagnostic actionable. Nous relions les capacités fondamentales et d'agent à travers un cadre de mesure au niveau de l'opération. Nous représentons chaque modèle avec un profil fondamental dérivé de références de capacité et un profil d'opération dérivé de trajectoires d'interaction, et définissons les lacunes de réalisation des capacités comme des écarts de performance observées par rapport aux prévisions basées sur les capacités fondamentales. Les exp
Machine translation (Meta M2M100); the English original follows.
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)
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.29751, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange