GDPx

GDPx is SemiAnalysis's general agentic benchmark, unveiled on November 2, 2026. It measures frontier models on 206 tasks spanning 11 domains that SemiAnalysis argues will capture the first trillion dollars of AI revenue — chip design, cybersecurity, biology, finance, legal, quant trading, recursive self-improvement, hardware design, materials science, alignment and software engineering — rather than weighting by share of GDP.

Tasks were built with 14 specialized RL-environment partners under a strict QA process. On the v1 leaderboard, Claude Opus 5.5 tops overall at 12.0%, ahead of Gemini 4 Carbon (10.9%) and GPT-6 Astra (9.6%), with no single model dominating every domain. SemiAnalysis plans updates every 3–6 months.

Design choices

• Domain selection mirrors SemiAnalysis's Tokenomics demand thesis, deliberately excluding lower-ROI sectors like government or insurance.
• Partners include Aria/Arcverex/Ergodex (chip), Arithmetic (cyber), Latch Bio, Living Context, Chakra, Third Moment, Extrapolate, Beautiful Machines, ValenceForge/Emulated, Fleet, Datacurve.
• Mixed binary/partial grading; pass@1 over three attempts; domains equally weighted.
• Some partner rollouts are open-sourced; most stay private to limit hill-climbing.

Why the low scores matter

A ceiling of ~12% is intentional: the suite aims to stay hard for GPT-6 / Opus 5.5-class systems so progress remains measurable.

1. Browse the live leaderboard and domain pages at https://gdpx.semianalysis.com/.
2. Read the methodology post on the SemiAnalysis newsletter for partner list and QA details.
3. Use domain breakdowns (not only the headline score) when comparing models for a specific vertical.
4. Expect the suite to refresh every few months; treat v1 numbers as a snapshot.