AI2 min
Research firm SemiAnalysis launches a hard AI agent test; the best model scores 12%
SemiAnalysis has launched GDPx, a benchmark for AI agents, meaning AI systems that carry out multi-step tasks on their own. It covers 11 fields, including chip design, cybersecurity, biology, finance, law, quant trading, software engineering, materials science and alignment. Anthropic's Opus 5.5 has the top score, at 12%.
Hype check: A 12% top score suggests today's agents are still far from doing expert work in these fields on their own. This is SemiAnalysis's own test and its own results, so treat it as one view.