Artificial Analysis has launched Intelligence Index v5, its most significant update since January. The new index is 60% built on private datasets to prevent contamination, featuring two new benchmarks: Frontier Code 2.0 and Terminal-Bench-Science. Upgraded components include HLE-Diamond, AA-Omniscience v1.1, and Crit Pt v2.
Claude Opus 5.5 leads the rankings with a score of 58, followed by GPT-6 Astra and GPT-6.1 Sol. The gap between proprietary and open-weight models has widened to 12 points. Cost per task varies dramatically, with GPT-6.1 Terra being the cheapest at $0.96 and Claude Fable 5.1 the most expensive at $13.04.
The Artificial Analysis Progress Index v1 was also introduced to track model capability over time.
Source: Hacker News · Summarized by HeadlinesBriefing