What happened
SemiAnalysis, a research firm, published GDPx version 1.0 with a public leaderboard. It covers 11 domains: chip design, quant trading, hardware design, cybersecurity, biology, finance, legal, software engineering, materials science, alignment and recursive self-improvement. The 206 tasks come from 14 specialized partners, mostly startups that build reinforcement-learning environments, and run inside simulated work environments rather than as single-file prompts.
SemiAnalysis argues that frontier models will earn most revenue in a few high-value sectors, so a benchmark should focus there rather than on the whole economy. It positions GDPx as a successor to OpenAI's GDPval.
The results, as measured by SemiAnalysis
On the GDPx leaderboard, Claude Opus 5.5 scores 12.0%, Gemini 4 Carbon 10.9% and GPT-6 Astra 9.6%. SemiAnalysis calls Opus 5.5 the best all-around model, winning quant trading, hardware design and chip design. It rates Carbon the best biology model, likely due to weaker safety classifiers, and narrowly ahead in software engineering, and Astra best in recursive self-improvement and materials science.
The authors stress that Opus 5.5 and Astra are held back by their safety classifiers in cybersecurity, biology and alignment tasks, and publish an alternative ranking that excludes those three domains. In finance and legal, every model scores 0% under binary grading, so the leaderboard reports average partial scores; GDPx uses a mix of binary and partial grading depending on the domain.
What it means for you
These are SemiAnalysis's own runs on its own task set, not independent replications. The overall scores sit close together, so the per-domain results are more useful than the headline ranking: check the domain closest to your work.
If your work touches security or biology, the classifier effect matters as much as raw capability: a model that refuses a legitimate task scores zero on it. Test candidate models on your own tasks before switching.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further