SemiAnalysis launches GDPx agent benchmark: the best model, Claude Opus 5.5, scores just 12%

SemiAnalysis published GDPx, a benchmark of 206 long agentic tasks across 11 domains it expects to capture the first trillion dollars of AI revenue, from chip design and cybersecurity to finance, legal, quant trading and recursive self-improvement, built with 14 data partners. Anthropic's Claude Opus 5.5 leads at 12.0%, ahead of Google's not-yet-released Gemini 4 Carbon (10.9%, at about a third of Opus's cost per task) and OpenAI's GPT-6 Astra (9.6%); Grok 4.7 scores 3.9%. Finance and legal tasks are so hard that basically every model scores zero, and SemiAnalysis says safety classifiers drag down Opus and Astra in cyber, biology and alignment. Dylan Patel: 'Official apology to Google from me, they cooked with Carbon.'