CLAIM-2026-08-15-03

Meta's own post announcing Muse Glimmer does not state the model's training compute or training token count.

Why it matters

Without the base model's training compute you cannot evaluate the one-third criterion directly, so the fixed fallback governs. The missing number is the hinge of the whole edition.

Sources

• metaairesea2026introduci
• meta2026muse