New GDPx Benchmark Finds Top AI Model Scores Just 12%

AI research firm SemiAnalysis unveiled GDPx on November 2, 2026. The rigorous new agentic AI benchmark spans 206 tasks across 11 high-value professional domains, including chip design, cybersecurity, biology, finance, legal work, quantitative trading, and software engineering.

The headline result: even the top-performing AI model scored just 12%.

The inaugural GDPx v1.0 leaderboard is led by:

• Claude Opus 5.5 by Anthropic: 12.0%
• Gemini 4 Carbon by Google: 10.9%
• GPT-6 Astra by OpenAI: 9.6%

Built with 14 specialized data partner startups, GDPx evaluates AI agents on long, realistic work tasks rather than simple question-and-answer formats. Each task has a default time limit of eight hours.

SemiAnalysis noted that the low scores for Opus 5.5 and GPT-6 Astra are partly attributable to their own safety classifiers, which can limit performance in sensitive fields such as cybersecurity and biology. The firm designed GDPx to focus on sectors it believes will generate the first trillion dollars of AI revenue, rather than weighting domains according to standard economic measures.

SemiAnalysis plans to update the benchmark every three to six months as AI models evolve.

Source: Unite.AI — SemiAnalysis Launches GDPx Benchmark Spanning 11 Agentic Domains