Artificial Analysis schedules Intelligence Index v5 for late October

ArtificialAnlys announced that version 5 of its Intelligence Index, a composite score used to compare AI models, will arrive in late October. The post calls it a significant leap in benchmarking the frontier and names two changes: adding Terminal-Bench Science, and a new coding benchmark built on a private dataset.

The post gives no scores or full list of changes, and says only to stay tuned. It does not state why the coding benchmark's dataset is private; test sets are typically kept private so models cannot have trained on them, but that motive is a general inference, not something the post says.