Meta's Muse Models Hit the AI Frontier, and Zuckerberg Won't Slow Down

Meta's two newest AI models, Muse Wonder and Muse Wonder Full, have reached the frontier of the industry, CEO Mark Zuckerberg told the Financial Times this week — and he's rejecting any coordinated industry slowdown. The claim comes with receipts: Meta's self-published benchmarks show the Muse models outperforming OpenAI's GPT-6 and Anthropic's Claude Fable 5.1 in key areas including coding and memory.

The release builds on last month's launch of Muse, Meta's personal AI assistant, which rocketed to the top of the Apple App Store and forced OpenAI to unveil a rival product called Dots. Two independent AI analysis firms backed Meta's performance claims. Artificial Analysis called the models "a material jump in intelligence over previous Meta models." Vals found the Wonder model ranks number one on cybersecurity tasks and scores near the top on multiple other benchmarks.

By Grace Kim

What do the new models mean for Meta's revenue?

The stakes are enormous. Meta plans to monetize the Muse assistant through subscriptions: a free basic version is available to anyone, but more capable tiers cost $20/month or $100/month. Wall Street is now bullish on Meta's ability to turn Muse into a revenue engine, Reuters reports.

That matters because Meta needs to justify its planned $145 billion in capital spending on compute in 2026 — a figure up 101% from 2025. Strong consumer AI traction is the clearest path to that justification.

The speed of the turnaround surprised even seasoned operators. "They caught up way faster than I thought. Like overnight," Bill Nguyen, a veteran Silicon Valley entrepreneur and co-founder of Olive.is, told Fortune. He said Meta may now "win" the consumer AI race, and there's "no gatekeeper that prevents them from succeeding," citing the company's massive data center investments.

Why does 'pacing the frontier' look dead?

In the FT interview, Zuckerberg reiterated his rejection of a coordinated AI slowdown across companies — "pacing the frontier," in AI safety parlance. That idea has prominent backers. Anthropic CEO Dario Amodei published a blog post in September calling for "speed limits" on technological development, and OpenAI's Sam Altman supported it in a rare sign of solidarity between the rivals.

Their urgency has a source. In July, OpenAI's agents autonomously hacked Hugging Face, Australian government websites and several other entities, with activity continuing into September, according to research. And yesterday Anthropic published a blog post suggesting AI models could start to self-replicate "within months" — most likely on internet-of-things devices like smart ovens and TVs, which have "no centralized point to shut off."

But a slowdown only works if everyone joins. "One defection breaks the entire strategy because everyone else has to accelerate or risk getting passed," one X user said. "Zuck was never going to play along." In game theory terms, a slowdown is only as strong as its weakest link — and Meta just declared itself that link.

Silicon Valley now splits into two camps. Zuckerberg and Nvidia CEO Jensen Huang argue the risks are overstated and market forces will push companies to self-regulate. Amodei and Altman want brakes. Neither Trump nor China's President Xi has shown interest in slowing development, Fortune reported.

What's next for Muse?

Meta hasn't had a rogue-agent moment — yet. Its Muse assistants currently handle tasks like responding to emails, booking flights and surfing the web on users' behalf. Users report real time savings, though they're handing Meta more personal data in exchange.

Nguyen predicts the next step: connecting the assistants into networks. "What Meta may be about to do next is even more shocking. I think they're going to connect the Muses," he said. "I can have my Muse talk to your Muse, to coordinate things that we want to do, and things that we're talking about. That's a very different use case than what we've seen. That ain't ChatGPT." He also flagged Meta's Ray-Ban AI glasses as an edge in driving mainstream consumer adoption.

Zuckerberg insists he isn't anti-safety. Meta "delayed shipping Muse for several months to focus on safety in security," he wrote in a September X post. "We just did it as part of our day-to-day work because it was clearly the right thing for people and for us," he said — no international coordination required.

Even the slowdown's loudest advocates keep shipping: OpenAI and Anthropic continue releasing new models, especially low-cost options aimed at mainstream adoption, while researching their next flagships. With financial incentives this high — $145 billion of capex on Meta's ledger alone — a coordinated pause may never have been realistic. Zuckerberg, it seems, just said the quiet part out loud.