SemiAnalysis launches GDPx, an agentic benchmark on which the best model scores 12%

SemiAnalysis released GDPx on November 2. The benchmark measures general agentic capabilities in 11 professional domains: chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering. The authors, Max Kan, Andrew Megalaa and Dylan Patel, argue that the first trillion dollars of AI revenue will be concentrated in sectors where frontier models can be deployed at high return. For that reason they chose not to weight tasks by share of GDP or by occupational taxonomies, and they deliberately left out sectors such as government, real estate and insurance. According to Unite.AI, version 1.0 contains 206 tasks built with 14 specialized data partners and ranks 13 models. Claude Opus 5.5 leads with 12.0%, at an average cost of $51.41 per task. Gemini 4 Carbon is second with 10.9% at $16.29 per task, and GPT-6 Astra is third with 9.6% at $53.99. Meta's Muse Wonder Full scores 9.0%, and Z.ai's GLM 5.3 ranks 13th with 2.7%. The headline score is each model's pass@1 result averaged over three attempts, with every domain weighted equally. SemiAnalysis notes that the safety classifiers on Opus 5.5 and GPT-6 Astra lower their overall scores, which affects how the cyber, biology and alignment results should be read. Because even the top model scores so low, the benchmark leaves a lot of room to measure future progress.