Meta has released Muse Wonder, which it calls its most capable artificial intelligence yet. Getting the best of it will cost money.
The company announced the Muse Wonder series on Oct. 12. It has two models: Muse Wonder and the larger Muse Wonder Full. Meta calls them "the next step on our scaling ladder and a major leap toward personal superintelligence." They succeed the Muse Spark models and were trained at Prometheus, Meta's gigawatt-scale data center.
Who gets what
Muse Wonder is available now to developers through the Meta Model API and Muse Code, Meta's coding tool. It costs $4 per million input tokens and $20 per million output tokens, or $0.20 per million for cached input, and has a context window of about one million tokens.
Muse Wonder Full has no per-token price. Meta says it is "available exclusively by subscription in Muse Code," on the Power Usage plan and a new Maximum Usage plan with 80 times the limits of the entry-level Everyday Usage plan. Meta lists the plans at $50 a month and $200 a month.
Regular users will have to wait, and many will have to pay. Meta says Muse Wonder "will soon power experiences in Muse, and will roll out first to subscribers."
Memory and persistence
Meta is pitching Muse Wonder as an engine for agents: software that carries out tasks on your behalf rather than just answering questions. It highlights two traits.
"Active Memory" carries context about a user across conversations and pulls up the relevant details when needed. "Active Focus" keeps the model working on a hard goal: Meta says it tries alternative approaches, works through blockers, asks for clarification when needed and "checks in before taking important actions."
In demos Meta published, the model builds a running plan in a browser and across apps, creates personalized gifts, and prepares a wedding seating chart from a list of complex requirements. Meta says its evaluations also cover law, finance and health tasks, including law firm software company Harvey's Legal Agent Benchmark.
For developers, Meta's main claim is coding. Meta says Muse Wonder Full "tops coding benchmark leaderboards, including DeepSWE and SWE-Marathon in independent third-party runs."
How it stacks up
Independent measures are more mixed. Muse Wonder Full scores 56 on the Artificial Analysis Intelligence Index, run at its maximum reasoning setting, level with Anthropic's Claude Sonnet 5.5 and behind Claude Opus 5.5 at 58, according to the index's leaderboard. The cheaper Muse Wonder scores 52. That is up from 48 for its predecessor, Muse Spark 1.3, Artificial Analysis says.
That matters for a company that has spent heavily to catch up. Meta says the release follows a "ground-up overhaul" of its AI work that started 14 months ago.
Safety report pending
An agent that reads your messages, calendar and documents raises the stakes. Meta says it trained Muse Wonder to resist adversarial instructions and handle private information carefully. In its announcement, Meta calls Muse Wonder its "most aligned model to date" and says it offers its "strongest safeguards to date" against chemical, biological, radiological, nuclear and cyber misuse.
Those claims can't yet be checked independently. Meta says a full report with additional safety evaluations "will be available soon."
What's next
Meta says it will use Muse Wonder to develop "specialized models that we can bring to everyone using Muse." For now, its strongest AI is a paid product, and the next test is how many of its billions of users will pay for it.