Artificial Analysis

Artificial Analysis is an independent benchmarking organization that takes the opposite approach to crowdsourced leaderboards: it runs models through a fixed, weighted basket of graded evaluations and produces a single Intelligence Index score from 0 to 100. Index v4.3.2 combines ten evaluations — including Terminal-Bench 4.0, Humanity's Last Exam, SciCode, GDPval-AA, and its own AA-Briefcase and AA-Omniscience suites — weighted across agents (30%), coding (20%), scientific reasoning (20%), and general (30%), with private never-before-published test sets at 45% weight to resist contamination.

Its model pages track 225+ models with intelligence, coding, and agentic indices plus price-per-task economics, and it has become the reference procurement benchmark alongside LMArena — its September 2026 snapshot had Claude Opus 5.5 leading at ~58, with Xiaomi's MiMo-V2.6-Pro the top open-weights model at 46. Public comparison pages are free; the business runs on enterprise benchmarking-insights subscriptions and private custom benchmarking work.

Limitation: unlike LMArena it has disclosed no public funding round, so its long-term independence rests on subscription revenue — and its scores are English-text-centric, with image, speech, and multilingual performance benchmarked separately. No public price list exists for the paid data platform.

Browse public leaderboards and model pages free at artificialanalysis.ai; contact them for enterprise benchmarking subscriptions.