Answer
Teams run their model on well known benchmark datasets and compare its scores against published results from other models to understand how competitive it is.
Real-world example A research team benchmarks their new language model against other popular models using a standard set of reasoning questions.