The Decoder· Manuel Uth·· 4 hours agoSelectedAI score67
SemiAnalysis Releases Real-World Work Benchmark GDPx; No AI Agent Scores Above 12%
No AI agent scores above 12 percent on SemiAnalysis' new real-world work benchmark
AI guide
Research firm SemiAnalysis has unveiled GDPx, an AI agent benchmark set in realistic simulated environments. Spanning 206 tasks across 11 domains—including chip design and quantitative trading—no model has yet scored above 12%.
Why it matters
By subjecting models to up to eight hours of complex, interactive professional workflows, the benchmark highlights the low completion rates of frontier agents in serious work scenarios.
Source:The Decoder · the-decoder.com