12% of the Economy: The Benchmark That Grades AI Agents Like Employees
On November 2, 2026, the research firm SemiAnalysis published GDPx, a benchmark that asks frontier AI models to do actual professional work — not answer questions about it, but do it, inside a simulated workplace, for up to eight hours per task. The best model in the world, Anthropic's Claude Opus 5.5, finished 12.0% of the tasks. Google's Gemini 4 Carbon came second at 10.9%. OpenAI's GPT-6…
On November 2, 2026, the research firm SemiAnalysis introduced GDPx, a benchmark that tasks state-of-the-art AI models with performing real-world professional duties within a simulated workplace, for up to eight hours per assignment. Out of the top three performers - Anthropic's Claude Opus 5.5, Google's Gemini 4 Carbon, and OpenAI's GPT-6 Astra - all scored below 12%.
The leading open-weight model, GLM 5.3, achieved a mere 2.7%. This 12% figure isn't indicative of models' regression; it signifies that the test has finally become authentic. For years, benchmarks have primarily focused on grading answers. GDPx, however, grades jobs, an entity with characteristics that exams do not. While one can perform well in an exam, it doesn't guarantee successful completion of a task, especially when the desired outcome is wrong, a rule is broken, or crucial files are missing.
This article delves into what GDPx measures, why the numbers are significantly low, and what these shortcomings imply for the deployment of AI agents in professional settings.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
