Meta Launches Muse Wonder and Muse Wonder Full, Says It Has Reached the Frontier

Meta introduced Muse Wonder and Muse Wonder Full, which it calls its most capable models to date, built for personal and professional agentic tasks. Meta says the pair shows its rebuilt AI organization has reached the frontier, and Muse Wonder is available through the Meta Model API and Muse Code today.

The announcement does not include a single benchmark score. That is the first thing I look for, and it is the part that is missing.

The claim, and what backs it

Meta's post puts the timeline plainly: "Fourteen months ago, we began an overhaul of our AI efforts, and in that time we’ve gone from rebuilding our foundations to pushing the frontier." Bloomberg Tech headlined its coverage that Meta says the new Muse model reaches the frontier after a 14-month push. The Information framed the launch as models meant to challenge OpenAI and Anthropic.

The Verge noted that Muse Spark was the first model from Meta's revamped Superintelligence Labs division, and that Meta now claims two new models that can go toe-to-toe with the industry's best. The Financial Times, citing an interview, reported that Mark Zuckerberg says Meta has caught up with rivals at the frontier and rejects calls for a slowdown.

Meta's announcement provides no score against a named competitor. Its post says the models offer capability in "law, finance, and health" and "frontier coding performance." Those are domains, not measurements.

The open questions are the ordinary ones: benchmark results, technical specifications, and parameter counts, none of which the announcement provides. A claim of catching up with frontier rivals is a comparison, and a comparison needs two columns.

What the models are said to do

Meta highlights two features. Active Memory is "the ability to carry what matters to you across conversations and bring back the details that are relevant to the task at hand." Active Focus is the ability to keep working when a task gets hard, try other approaches, recover from errors, and follow through, asking for clarification and checking in before doing anything important.

That second feature is the more interesting engineering claim, because a stuck step, a wrong turn or an irreversible action is precisely where I would want an agent to check itself. Whether Active Focus recovers from errors better than the competition is an empirical question. Meta's post does not report a failure rate, so it stays a description of intent.

On safety, Meta says it trained the Muse Wonder family to avoid sharing private information without permission, turn down requests to assist in actually harmful acts, and strengthen defenses against cyber attacks. The models are trained, evaluated and released under its Meta Super Intelligence Scaling Framework. A full report with additional safety evaluations "will be available soon." Soon is not a date, and the announcement gives none.

Price, access and a small discrepancy

The one hard set of numbers comes from outside Meta. According to OpenRouter's listing, Muse Wonder has a context window of 1,048,576 tokens, accepts text, images, video, audio and PDF documents, and returns text. It costs $4.00 per million prompt tokens and $20.00 per million completion tokens.

Meta's announcement does not detail pricing or API terms for Muse Wonder Full.

Availability is where the accounts diverge. Meta's post says Muse Wonder is available today through the Meta Model API and Muse Code, and that it "will soon power more of Muse, rolling out first to subscribers." CNBC reports that Muse Wonder is already available to Muse app subscribers, and that Wonder Full is available to Muse Code subscribers. Meta's post does not spell out release channels for Wonder Full.

What it means, and what to watch

The strategic read is straightforward. Meta spent 14 months rebuilding and wants the market to know the rebuild is finished. Wonder is also the engine for the Muse agent: The Verge reports Meta plans to power Muse with Wonder, starting with subscribers, and to develop specialized models it "can bring to everyone using Muse." The launch is partly a capability claim and partly a product plumbing announcement.

What to watch:

• Independent evals. Look for third-party results that test Meta's frontier claim against OpenAI and Anthropic models.
• The safety report. Meta promised additional safety evaluations "soon." Watch for the date and the contents.
• Wonder Full's terms. Price, API access and who can actually use it.
• The Muse rollout. The cited Verge report gives no rollout date for the specialized models intended for all Muse users.

The launch is real. The models exist, the price is listed, and the pitch is clear. The frontier claim, for now, is a press release's worth of evidence. Show me the eval, not the vibe.