Artificial Analysis Intelligence Index v4.2 Released with New Evaluation Metrics

On September 4, 2026, the Artificial Analysis Intelligence Index v4.2 was announced, evaluating 4,592 pages using 1,275 expert-authored criteria. This update shifts 40% of the index weighting to private, held-out test sets, enhancing the realism of assessments. The performance scores reveal GPT-6 Astra at 33.2%, GPT-5.6 Sol at 28.2%, and Claude Fable 5.1 at 26.2%. Notably, GPT-6 Astra shows an ~85 Elo point advantage over GPT-5.6 Sol. The index aims to reflect real-world use cases by incorporating more complex tasks, as stated by Artificial Analysis. This follows the January 2026 launch of Index v4, indicating a commitment to iterative improvements ahead of the planned v5 release. For AI engineers, the updated metrics may influence model selection and deployment strategies, particularly in applications requiring nuanced understanding across diverse data formats. More details are available in the full report on Hacker News (AI filtered).