ML//benchmark//GPQA

GPQA (Graduate-Level Google-Proof Q&A) is a benchmark of questions so hard that PhD domain experts score about 65% even with internet access. It is used alongside MMLU Pro to measure deep expert-level reasoning, and its hardest subset, GPQA Diamond, is the standard metric frontier labs report.

GPQA (Graduate-Level Google-Proof Q&A) is a benchmark of questions so hard that PhD domain experts score about 65% even with internet access. It is used alongside MMLU Pro to measure deep expert-level reasoning, and its hardest subset, GPQA Diamond, is the standard metric frontier labs report.