According to CoinMeta, Artificial Analysis has updated the intelligence ranking to v4.2, increasing the weight of private tests from 20% to 40% in order to reduce the possibility of targeted score boosting. The new version includes aa-briefcase and gdp.pdf; the former simulates real knowledge work projects, while the latter requires the model to process 4592 pages of professional pdf. gpqa and diamond have been removed from the comprehensive index. gpt-6, astra, and gemini achieved high scores of 96.3% and 95.3% respectively. flash still ranks first with 5.1 points, followed by claude and fable in second place with 5.1 points each, which is 4 points higher than gpt-5.6 and SOL. artificial and analysis indicate that astra uses fewer token outputs among models of the same level.