Artificial Analysis reweights its Capability Indices
Artificial Analysis has released Capability Indices v1.1, adding agentic tool-use and document benchmarks to its six domain scores and dropping a customer-service test, which changes how models are ranked for professional work.
Artificial Analysis announced version 1.1 of its Capability Indices. The indices score models in six domains: Finance & Accounting, Strategy & Ops, Legal, Healthcare & Medical, Engineering and Economics. Each index maps tasks from O*NET occupations to benchmarks and weights each benchmark by how often its capability appears across those tasks. The update uses domain-specific slices of evaluations from Intelligence Index v4.2 and v4.3 where they exist.
All six indices now use AA-Briefcase alongside GDPval-AA v2 for agentic knowledge work. Four indices drop Agentic Customer Interaction (𝜏³-Banking) and add Agentic Tool Use from AutomationBench-AA, drawing on the slice that fits each domain. Tool use now accounts for 30% of Strategy & Ops, 10% of Finance & Accounting, 10% of Healthcare & Medical and 5% of Legal. Finance, Strategy & Ops and Legal also add GDP.pdf to their long-context component.
Healthcare & Medical adds MLCR-AA at 15%, which tests reasoning across long clinical records. To make room, the index lowers the weights for medical knowledge, non-hallucination and reasoning. Engineering moves from Terminal-Bench v2.1 to v4.0 and raises terminal use from 5% to 15%. It also removes GPQA Diamond from its reasoning component. Economics raises agentic knowledge work from 15% to 25% and lowers long-context reasoning from 15% to 5%.
The announcement does not say how the changes affect model rankings. Per-domain results and the methodology are on the Artificial Analysis site.