Artificial Analysis Intelligence Index v4.2 (artificialanalysis.ai)

On September 4, 2026, the Artificial Analysis Intelligence Index released version 4.2, introducing significant updates aimed at enhancing the evaluation of AI models through more rigorous and realistic assessments. Key additions include AA-Briefcase, which evaluates models on complex multi-week knowledge work tasks through a private test set, and GDP.pdf, a challenging document reasoning test that requires models to synthesize information from a vast 4,592-page collection. This version also places increased emphasis on held-out test sets—now constituting 40% of the Index's weight—to reduce the risk of models gaming the assessments, while improvements to grading infrastructure aim to enhance scoring reliability. This release is pivotal for the AI/ML community as it sets a higher standard for benchmarking, ensuring models demonstrate true agentic capabilities in knowledge work, which reflects real-world applications more accurately. With leading models like Anthropic’s Claude Fable 5.1 and OpenAI’s GPT-6 Astra emerging at the top of the leaderboard, the update highlights increased competition among labs and signals a shift towards more practical evaluations in AI assessment. As the development of version 5 progresses, users can anticipate further refinements aimed at maintaining the Index’s relevance in a rapidly evolving technological landscape.