GPT-5.6 Sol Score Corrected to 83%

scaling01 · x · 2026-07-10

AcerFur stated that GPT-5.6 Sol's score on FrontierMath T4 v2 has been corrected to 83%. The shared post added that the author finds GPT-5.6 Pro to be even better, citing the team's explanation: previous blog scores originated from pre-release evaluations and have since been re-run on public models, yielding results more consistent with overall trends.

Original post →

More from Models

Models channel →