According to CoinMeta, SemiAnalysis has collaborated with 14 professional data companies to launch GDpx, which is specifically designed to test the actual working capabilities of top-tier AI models. The evaluation suite includes 206 high-difficulty tasks covering 11 fields such as chip design, finance, law, biology, cybersecurity, and software engineering. Researchers created a complete professional environment for the AI models, replicating a real acquisition transaction worth $880 million in financial tests. They provided approximately 4,000 anonymous documents for the models to review contracts, conduct accounting checks, and modify financial models on their own. The results showed that Opus won with a comprehensive score of 12 out of 5.5, followed by Google Gemini with 4 points, and Carbon with 10.9 points. The finance and law fields were particularly challenging, with all participating models scoring zero in these areas. Although the top models possess individual capabilities, they struggled to meet multiple requirements in complex tasks.