Benchmark
GDPval
Evaluate AI on professional deliverables across major industries.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
1,320 tasks spanning 44 occupations and nine US industries; the public gold subset contains 220 tasks.
Methodology
Models produce documents, spreadsheets, presentations, and other deliverables from realistic briefs and reference files. Domain experts compare the resulting work.