Benchmark
GDPx
Compare how well AI agents complete specialist work, from designing chips to analyzing financial records, and how much each attempt costs.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
SemiAnalysis and 14 specialist partners contributed 206 tasks across 11 domains. Each task places the agent in a working environment.
Methodology
The default grade averages three attempts per task. Depending on the task, a grader awards partial credit or records success or failure. Partners carry equal weight within each domain, and each domain contributes equally to the overall score. The team chose the domains for their expected exposure to AI. The aggregate measures performance on that selection.