TL;DR
Capability indices combine benchmark evidence using domain-specific weights. Read the occupational mapping and version changelog before interpreting a rank, because a changed score can reflect a changed measurement as well as a changed model. This reading note explains why a weighted capability index is not a measured job-completion rate or a complete description of a model's strengths.
What to look for
The indices connect occupational tasks from O*NET to benchmark capabilities. Read the domain changelogs and weight tables to understand which kinds of work contribute to a score, rather than treating one overall rank as a complete model profile.
Keep in mind
These are weighted benchmark indices, not measured job-completion rates. Changes between index versions can reflect different evaluations and weights as well as changes in model capability.
Read the original work
Announcing Artificial Analysis Capability Indices v1.1 ↗By Artificial Analysis · Artificial Analysis
Published Sep 14, 2026. Source checked Oct 5, 2026.
Reading note by West Mao and Alex George · Added Oct 5, 2026. This page is a short editorial introduction; the full article belongs to its original publisher.
