GDPx: Measuring Frontier Model Capabilities Across the Most Valuable Domains (newsletter.semianalysis.com)

GDPx has been launched as a new benchmark designed to assess the capabilities of frontier AI models across critical domains, including chip design, cybersecurity, biology, finance, and legal fields. Unlike previous benchmarks, GDPx emphasizes evaluating general agentic capabilities specifically within high-impact sectors, rather than broadly measuring overall economic contributions. The current frontier-level benchmark reveals that even the top models like Opus 5.5 and GPT-6 Astra score only 12%, indicating the challenging nature of tasks they face. The benchmark deliberately omits industries such as government and real estate, focusing instead on where AI can deliver the most significant returns on investment. The significance of GDPx lies in its potential to set a new standard for AI evaluation, particularly in creating realistic task environments that mirror real-world complexities, unlike previous benchmarks such as OpenAI’s GDPval. By utilizing specialized RL companies to provide rich, domain-specific tasks, GDPx overcomes many limitations of existing evaluation frameworks that rely on simplistic or unfair test designs. This initiative highlights the importance of safety classifiers and domain expertise in selecting effective AI models, ultimately paving the way for a more nuanced understanding of AI capabilities in applied scenarios. As models like Gemini 4 Carbon show promise, the AI community is keenly interested in how GDPx will influence future developments and applications of AI technology.