Meta released Muse Spark 1.3 on 2 September through Muse Code and the Meta Model API. Within hours the model was being ranked against the frontier on a score that, according to Meta's own launch post, belongs to a configuration nobody outside Meta can currently run.
The sentence that undoes the leaderboard
Meta's post states the position plainly: Muse Spark 1.3 "is rolling out today in Muse Code and Meta Model API," and "previously available reasoning modes are available today with max reasoning coming shortly after we finish additional safety testing." The top reasoning setting is not late documentation or a staged regional rollout. It is explicitly gated behind unfinished safety work, on a model that is otherwise generally available.
Why the distinction changes the ranking
Reasoning-mode selection is the single largest lever on a modern model's benchmark position, and it is also the largest lever on its cost. A score produced at maximum reasoning and a score produced at the tier below are not two measurements of the same product; they are measurements of two different price points, only one of which is purchasable today. Comparison tables that place Muse Spark 1.3 against rivals on the higher figure are comparing a shipped competitor with an unshipped configuration.
What the common framing gets wrong
The launch is being written up as "Meta ships a frontier-class model." What Meta shipped is the model at its already-available reasoning tiers, plus a commitment to release the top tier later. The gap is not pedantry: the entire argument for the release being competitive rests on the number attached to the part that did not go out. Second correction to make: this is an API and product release, not an open-weights drop, whatever the surrounding coverage of Meta's open-source posture implies.
The pattern this fits
Holding the strongest configuration back for additional safety testing while publishing its evaluation results is becoming a recognisable release shape across several labs. It lets the scoreboard reflect the ceiling while the shipping product sits below it, and it puts the burden on readers to notice which number applies to what they can actually call. Meta disclosed the split in the post itself, which is more than the aggregation around it preserved.