Unsteady Metrics and Benchmarking Cultures of AI Model Builders

Analysis of benchmark selection practices in AI model builder press releases and blog posts, introducing Benchmarking-Cultures-25 dataset to examine how companies choose which benchmarks to highlight and what this reveals about state-of-the-art claims.