Artificial Analysis Intelligence Index v5 is a major overhaul of the composite benchmark measuring AI model capabilities across real-world tasks. It adds two new private benchmarks—FrontierCode 2.0 for coding agents and Terminal-Bench-Science for scientific research—upgrades three existing evaluations, and shifts 60% of the index weight to private datasets or held-out answers. The company also launched Progress Index v1, which tracks language model capability changes over time on a single scale using item response theory.
Artificial Analysis Intelligence Index v5 is a composite evaluation of language models across knowledge work, coding, science, and general knowledge. Its nine evaluations include both agentic tasks and other forms of model assessment; the index assigns 25% of its weight to each of the four categories. Version 5 introduced FrontierCode 2.0 and Terminal-Bench-Science, upgraded several existing evaluations, and uses private test sets or held-out answers for 60% of the index. Artificial Analysis presents the index as a comparative synthesis, while noting that it may not apply directly to every use case.
