Artificial Analysis has released its Intelligence Index v5.0.1, evaluating Claude Fable 5.2 (max) across nine benchmarks including AA-Briefcase v1.1, Automation Bench-AA, and Terminal-Bench 4.0. The index measures both raw intelligence and cost-efficiency, calculating a weighted average cost per task based on input, cache, and reasoning token prices.
The model is categorized as having no publicly available API, and its commercial use is restricted. Performance is scored across capability indexes such as agentic software engineering, legal reasoning, and medical long-context tasks.
Evaluations like AA-Briefcase Elo aggregate rubric pass rates and analytical quality, while others like HLE-Diamond and Crit Pt v2 test hallucination-gated accuracy and scientific terminal workflows. The full methodology details how each benchmark is weighted and run.
Source: Hacker News · Summarized by HeadlinesBriefing