Artificial intelligence / news brief
SemiAnalysis launches GDPx benchmark for long-horizon AI agents
- Event
- Published

SemiAnalysis launched GDPx, a 206-task benchmark built with 14 partners across 11 technical and professional domains. Its leaderboard averages three attempts per task and puts Claude Opus 5.5 first at 12.0%, but the authors say safety refusals depress some domain scores and the results have not been independently reproduced.
Why it matters
GDPx adds tool-using, multi-hour environments to agent evaluation, while its partner-built tasks and self-run leaderboard require external audit before broad conclusions.
Sources
- GDPx: Measuring Frontier Model Capabilities Across the Most Valuable Domains (external site)SemiAnalysis · primary source ·
- SemiAnalysis unveils GDPx benchmark, and the best AI model scores 12% (external site)Crypto Briefing · coverage ·
- SemiAnalysis Launches GDPx; Best Frontier Model Hits 12% (external site)AI Weekly · coverage ·