I Checked 30 Frontier Model Cards. Here Are the Benchmarks Labs Report

A survey of 30 frontier model cards reveals that labs report a narrow set of benchmarks, often omitting reproducibility details and process-oriented metrics. The analysis highlights a growing gap between headline accuracy scores and real-world reliability, urging evaluators to demand more transparent and comprehensive reporting.
Model comparisons based only on aggregate accuracy can miss inefficient search, unsupported answers, and silent failures.