SemiAnalysis launches GDPx agentic benchmark across 11 high-value domains
On 2 Nov 2026 SemiAnalysis launched GDPx covering 11 domains and 206 tasks (chip design, cyber, bio, finance, and more). The public leaderboard shows Claude Opus 5.5 at 12.0, Gemini 4 Carbon 10.9, and GPT-6 Astra 9.6—underscoring how hard these tasks remain even for top models.
Verified on the GDPx site and the same-day SemiAnalysis newsletter (dated Nov 02, 2026): SemiAnalysis released GDPx, an agentic benchmark aimed at domains it sees as highest-ROI for frontier models, rather than a GDP-share or SOC/O*NET-wide survey.
The site lists 11 domains, 14 partners, and 206 tasks spanning chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement (RSI), hardware design, materials science, alignment, and software engineering. The newsletter calls it a true frontier-level benchmark: even the best model scores only about 12%.
On the site’s domain-compare table (as fetched today), the top three overall scores are Claude Opus 5.5 12.0, Gemini 4 Carbon 10.9, and GPT-6 Astra 9.6. The newsletter also notes safety classifiers can sandbag some models on cyber/bio/alignment, and that finance and legal tasks are so hard that binary grading yields near-zero for essentially every model. Figures are SemiAnalysis’s; this item does not claim an independent IdeaDawn re-run.