GDPx measures agentic capabilities across 11 high-value domains

What is changing?

SemiAnalysis has released GDPx to evaluate general-purpose AI agents in economically valuable specialist domains. The benchmark spans 11 areas, including chip and hardware design, cybersecurity, biology, finance, legal work, quantitative trading, recursive self-improvement, materials science, alignment and software engineering.

In the initial published results, the best model achieved an overall score of only 12%. The authors say safety classifiers restrict some models’ responses in cybersecurity, biology and alignment, and that rankings can change when those three areas are removed. The results therefore reflect not only raw capability but also whether a model is willing to answer a task.

GDPx offers technical teams a difficult, domain-focused comparison when selecting models. However, the benchmark deliberately excludes sectors such as government, real estate and insurance, so production decisions should still include an organization’s own tasks, costs and safety policies.