Tag
This paper introduces an expert-validated STEM QA dataset created to address gaps in existing benchmarks, demonstrating low performance from frontier AI models and potential for model training.