AA Intelligence Index under fire as Muse Spark 1.3 outranks GPT-6 Astra

_weiping · x · 2026-09-04

DanDr1s questions the credibility of the Artificial Analysis Intelligence Index after it ranked Muse Spark 1.3 (62) above GPT-6 Astra (61) — despite GPT-6 Astra scoring 98.6% on ARC-AGI-3, 97.6% on FrontierMath, and 100% on ExploitBench.

weiping adds that any single-number "benchmark" is inherently misleading, and benchmaxxing the index is worse than gaming any specific benchmark: it obscures key capabilities and leads people to dismiss meaningful benchmarks not included in the index.

Original post →

More from Models

Models channel →