Artificial Analysis introduced Intelligence Index v5 on October 27, changing the evaluation mix. Our Benchmarks page now uses a single v5 snapshot checked November 1; older article snapshots remain dated history.

The leading listed settings are Carbon high at 59, Opus 5.5 max with fallback at 58, and Astra max at 57. Sol max scores 55. These are composite index points, not percentages or universal quality rankings.

Kimi K3.1 Pro is 46 on v5. Its earlier 55 came from the older index; the change alone does not demonstrate a model regression. Moonshot's downloadable-weights promise remains November 4, while its products and API are already available.

Keep measurements separate

Reported reasoning effort and fallback settings are preserved on each row; unlisted settings remain unspecified. Existing Arena and Terminal-Bench figures are explicitly older snapshots, not newly measured scores. Our editorial category assessments remain separate and have not been recalibrated by copying AA numbers.