We ran Opus 4.8 and GPT-5.5 against our GDPVal+ dataset.

GDPval+ is a progressive curriculum of economically valuable workplace tasks, expert-authored across all 20 O*NET sectors and 100+ occupations.

Each datapoint pairs a professional prompt, reference files, a golden solution, and a weighted rubric, packaged as both a raw definition and a Harbor-ready evaluation. Our dataset delivers thousands of single-turn tasks averaging 5-10 hours of human completion time.