Artificial Analysis launches Intelligence Index v5, built mostly on private tests; open-weight gap hits 26 points
Artificial Analysis rebuilt its headline index so about 60% of the score comes from private or held-out tests, adding FrontierCode 2.0 and Terminal-Bench-Science and weighting four categories equally. Claude Opus 5.5 led at launch with 58, and the best downloadable-weight models scored 32. Google's Gemini 4 Carbon, announced a day later, claims 59.
Why it matters
A widely cited leaderboard has reset its scale. Comparisons against v4.x scores no longer hold, and the open-versus-closed gap now looks much larger.