Has Meta finally caught up? Inside the launch of Muse Wonder

After 14 months of rebuilding and billions spent on talent, Meta says its new Muse Wonder models can take on OpenAI and Anthropic.

For much of the past two years, Meta looked like the big tech company that had fallen behind in AI. On 12 October it set out to change that story, launching Muse Wonder and Muse Wonder Full, which it calls its most capable AI models yet.

What Meta released

There are two models. Muse Wonder will soon power Meta's Muse personal AI agent, starting with paying subscribers. The more powerful Muse Wonder Full is available to subscribers of Muse Code, Meta's AI tool for writing software. In a blog post, the company said: "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."

The launch follows a year in which Mark Zuckerberg pledged billions of dollars to AI and aggressively hired researchers away from rivals. Muse Wonder is the first real test of whether that spending has paid off.

Why subscriptions matter

Meta has always made its money from advertising. The Muse agent, which rolled out last month, gives it a second route: people paying directly for a personal AI that can help with tasks across its apps. If that works, Meta's huge AI spending could start to earn its keep. The catch is trust. A personal agent built by a company whose business is data will face hard questions about privacy, and Meta has been criticised on that front before.

Investors were cautious on day one. Meta's shares slipped 0.7% in early trading the morning after the launch, according to Reuters.

A crowded week at the top

Muse Wonder arrived in one of the busiest weeks the AI industry has seen, with Google's Gemini 4 Argon also landing close to the front of the pack. The gap between the leading labs is now small enough that each new release can reshuffle the rankings.

Our take

Catching up is not the same as pulling ahead. Meta has shown it can build a frontier-class model again, which seemed in doubt a year ago. The bigger question is whether ordinary people will pay a monthly fee for a Meta AI assistant. That, more than any benchmark, will decide whether this comeback is real.