SemiAnalysis has released GDPx, a benchmark designed to measure how well perform as autonomous agents across 11 specialized domains. The benchmark includes tasks in chip design, cybersecurity, hardware design, quantitative trading, finance, legal work, bioscience, software engineering, materials science, and AI alignment.
The benchmark tests models on practical, domain-specific problems that require reasoning and planning. According to the results, Anthropic's Opus 5.5 achieved the highest score at 12%, suggesting that current AI systems still have significant room for improvement on these complex, specialized tasks.
GDPx fills a gap in AI evaluation by focusing specifically on agentic capabilities rather than simple text generation or question-answering, offering a more realistic measure of how AI agents might perform in professional and technical roles.
This is an AI-written summary. The original reporting belongs to Techmeme. How we write these · Spot a mistake?