The base Muse Wonder model scored 52 on the Artificial Analysis index and is priced on Meta's API at $4 per million input tokens, $0.20 for cached input, and $20 per million output tokens across a 1M token context window. While those token rates match Opus 5.5 and are 60% below GPT-6 Astra and Fable 5.1, heavy input usage of 7.69M tokens per task brings Muse Wonder's average task cost on the benchmark to $4.52, compared with $3.26 for GPT-6 Astra and $5.98 for Opus 5.5. Meta has not released per-token pricing for Muse Wonder Full, which is restricted to paid Muse Code plans.
Meta shares rose as much as 4.7% following Monday's launch and gained another 3.56% to $769.40 on Tuesday as Truist analysts cited early Muse adoption. Chief AI Officer Alexandr Wang said late Monday that Meta fixed a launch bug that caused unusually fast quota depletion for Muse Code subscribers and reset user limits. In its model documentation, Meta noted that "visual and spatial reasoning, especially 3D tasks, remains a known limitation," and disclosed that the model captured 71 of 92 cybersecurity benchmark flags via unintended vulnerabilities while requiring safeguards to reduce chemical and biological risks to moderate or lower.