AI benchmarks

MaxxBench changes coding tasks to test whether AI follows new requirements

MaxxBench revises 89 DeepSWE tasks and finds that familiar coding scores can hide differences in adaptation. We examine the tests, costs and limits…