SemiAnalysis announced on Monday (November 2, 2026) the GDPx, a new benchmark designed to measure the agentic capabilities of frontier AI models in high-economic-value domains. Even the best-performing model, Claude Opus 5.5, achieved only 12% success.
What is GDPx?
Unlike general benchmarks based on GDP or occupational classifications, GDPx focuses on sectors where frontier intelligence is expected to generate the first trillion dollars in AI revenue, according to SemiAnalysis. The 11 domains are: chip design, cybersecurity, biology, finance, legal, quantitative trading, recursive self-improvement (RSI), hardware design, materials science, alignment, and software engineering.
Each domain had at least one specialized data and RL environment partner. There are 14 partners in total, with multiple for chip design and materials science. Tasks include realistic, context-rich environments such as virtual data rooms for financial transactions, simulated laboratories, and cybersecurity setups with running services — contrasting with older benchmarks like OpenAI's GDPval, which provide context more directly and with limitations.
Initial results
In the overall ranking, Claude Opus 5.5 leads with 12.0%. Next is Gemini 4 Carbon (not yet widely available) at 10.9%, and GPT-6 Astra lower. Models show performance spikes in specific domains: Opus in trading, hardware design and chips; Gemini Carbon in biology and software engineering; Astra in RSI and materials science. In finance and legal, nearly all models scored zero.
SemiAnalysis notes that safety classifiers affect scores in domains like cyber, biology and alignment. The benchmark also reports cost per task and output tokens, enabling cost-benefit analysis.
Why it matters
GDPx highlights how much remains for AI agents to reliably perform complex professional work. While many benchmarks saturate quickly, this one stays challenging, offering a useful metric to track real progress in high-impact areas. The collaborative approach with specialized partners aims to ensure task quality and realism.
Results and details are available on the official site and SemiAnalysis newsletter.
Sources
Image credit: SemiAnalysis / GDPx illustration
Transparency: This content was created, edited or reviewed with the aid of artificial intelligence. Information was cross-checked with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot
