What happened
SemiAnalysis published GDPx on November 2, 2026, with a dedicated leaderboard at gdpx.semianalysis.com and a blog write-up. The site lists 11 domains (including chip design, cybersecurity, biology, finance, legal, quant trading, RSI, hardware design, materials science, alignment, and software engineering), 14 specialized data partners, and 206 tasks. On the all-domains table, Claude Opus 5.5 scores 12.0, Gemini 4 Carbon 10.9, GPT-6 Astra 9.6, Muse Wonder Full 9.0, and Claude Fable 5.1 8.5. SemiAnalysis says Opus 5.5 and GPT-6 Astra overall scores are sandbagged by safety classifiers on cyber, biology, and alignment, and that Muse Wonder Full leads cyber by a meaningful margin. Finance and legal are described as so hard that models score essentially 0. The blog contrasts GDPx with OpenAI’s GDPval (September 25, 2025) as a knowledge-work suite that predated today’s agents.
Why it matters
People will quote “agents at 12%” without naming the test. GDPx is SemiAnalysis’s mix of high-ROI domains, not a GDP-share occupational sample. The useful habit is still: which named tasks, on which date, scored by whom — and whether a safety classifier blocked the run.
SemiAnalysis says excluding cyber, biology, and alignment changes the ranking because safety classifiers sometimes fire on Opus 5.5 and GPT-6 Astra.
The blog and site are SemiAnalysis’s; partner labs contributed tasks, and SemiAnalysis says it did final QA.