Independent max-configuration page recording an Intelligence Index score of 62, 120 million output tokens, and the explicit label that this mode is not publicly available.
Independent current measurements for the accessible xhigh configuration: Intelligence Index 61, published API price and speed, and 100 million output tokens across the evaluation.
Official evaluation methodology showing Meta used max reasoning for Muse Spark 1.3, xhigh for Muse Spark 1.2, and mixed the best available results from Meta runs, official leaderboards, or provider self-reports.
Official Sep. 2 release for current Muse Code and Meta Model API availability, max reasoning pending additional safety testing, Meta's agent/coding and efficiency claims, and the future open-weights roadmap.
Sensemaker daily brief distinguishing the Muse Spark 1.3 modes developers can use from max reasoning still under safety testing and an open-weights release still on Meta's roadmap.
The end
