What changed
SemiAnalysis launched GDPx on November 2, 2026, testing frontier AI models on agentic tasks across 11 economically valuable fields, including chip design, biology and finance.
The first leaderboard puts Anthropic’s Claude Opus 5.5 first at 12.0%, Google’s Gemini 4 Carbon second at 10.9%, and OpenAI’s GPT-6 Astra third at 9.6%. Gemini led biology and software engineering; GPT-6 Astra led recursive self-improvement and materials science.
Why it matters
GDPx is trying to measure useful work, not general cleverness. Its structure follows task-based evaluations such as SWE-bench, asking whether a model can complete defined practical jobs in fields where SemiAnalysis expects AI to produce the highest returns.
The striking result is how low every score remains. Even the leader completed only 12% of the tested tasks successfully. That makes the ranking useful for spotting relative strengths, but weak as evidence that any model can reliably run an entire professional workflow.
The domain choices also shape the result. SemiAnalysis selected areas it expects to generate strong returns, rather than weighting them by GDP share or occupational distribution. Government and insurance are excluded, so this is a focused map of commercial AI potential, not a complete measure of economic capability.
There is one important qualification: the tasks were built with 14 partner startups, while the validation process beyond those partners is not explained. That leaves room for the leaderboard to change as the benchmark adds industry feedback and new organizations. SemiAnalysis plans updates every three to six months, so the next versions may tell us as much about the benchmark’s design as about the models themselves.
Comments
No comments yet.