Artificial Analysis Index v5 puts most weight on private tests
Sorted AI29 OctArtificial Analysis has launched Intelligence Index v5, which it says puts 60% of the Index on evaluations with private test sets or held-out answers, including Cognition's FrontierCode 2.0 and a new Terminal-Bench-Science set. On its figures, Claude Opus 5.5 (max) leads at 58, ahead of GPT-6 Astra (max) at 57 and GPT-6.1 Sol (max) at 55, and the gap from the leader to the best open-weights model widens from 3 points in v4.3 to 12 points (Kimi K3.1 Pro at 46; among released weights, DeepSeek V4.1 Flash 1017 and GLM-5.3 lead at 32). Cost per Index task among leading models spans more than 13x on its chart, from about $0.96 for GPT-6.1 Terra (max) to $13.04 for Claude Fable 5.1 (max). Treat these as Artificial Analysis's own rankings.
Why it matters: If you pick models for coding or knowledge work, use v5 as one harder-to-game comparison, but still run your own tasks; the closed/open gap and cost spread are their figures, not a reason to switch overnight.
Artificial Analysis ↗AI-assisted, reviewed by Sorted AI
▲ 0
0 comments