Meta's Muse Wonder, at $20 to $200 a month, is framed as a test of its superintelligence team

Four plans carry the release. Muse app users pay $20 or $100 a month to reach Muse Wonder ^[3], and developers pay $50 or $200 a month on Muse Code to reach Muse Wonder Full ^[4], so the dearest plan costs ten times the cheapest ^[15]. Neither report includes an API price, benchmark scores or subscriber numbers.

Cryptobriefing called the release the first big report card for Meta Superintelligence Labs since its 2025 overhaul ^[7], and wrote that Bloomberg framed the launch as a test of the team's capabilities ^[6]. Mark Zuckerberg announced the lab on June 30, 2025 ^[10], and it had roughly 3,000 to 3,400 staff by late 2025 ^[11]. Wonder is not its first release: Muse Spark arrived in April 2026, and Muse Image and Muse Video followed in July ^[12].

In our view the subscription is a sharper test than any benchmark. A user who runs the free app on real tasks and then pays $20 a month has made a judgment no leaderboard records. The counter-case sits in CNBC's own report: the Muse app, which is free, reached the top of Apple's App Store after Meta introduced it last month ^[13], and that measures appetite for a free app. It does not show how many of those users will pay.

Wang described what the paid model is meant to do differently. He said: "We specifically tried to train the model to be able to try other approaches instead of stalling and then checking back in with the user at any time." ^[14] Asked about the message-access complaints, he said, "I think there are certainly instances users have authorized various communications or whatnot earlier in a conversation, and then may have forgotten about that," Wang said ^[18].

We'd expect an agent trained to continue past a setback to act more often without a fresh prompt, and each such action raises the odds that an earlier authorization is the one in dispute. That is our inference. The complaints concern the Muse app Meta introduced last month ^[13], before Wonder arrived.

We would be wrong about the subscription being the test if most paying customers are developers on Muse Code or the API. In that case the benchmarks and the API's price against OpenAI and Anthropic decide the result ^[5].