Meta has reached the “frontier” with the two new models it released this week, Muse Wonder and Muse Wonder Full, Zuckerberg told the Financial Times in an interview. They build on momentum from last month’s release of Muse, Meta’s personal AI assistant, which rocketed to the top of the Apple App Store, and compelled OpenAI to unveil a rival product, Dots. 

Meta’s self-published model benchmarks show its Muse models outperforming other frontier models, such as OpenAI’s GPT-6 and Anthropic’s Claude Fable 5.1, in key areas such as coding and memory. Two independent AI analysis companies also praised its performance. One firm, Artificial Analysis, called the models “a material jump in intelligence over previous Meta models,” and another, Vals, found that the Wonder model is number one on cybersecurity tasks, and scores near the top on multiple other benchmarks. Meta has entered the ring.

The new models will power future upgrades to the Muse assistant, which Meta plans to monetize through subscriptions. While anyone can access a free, basic version of Muse, more capable versions require a $20/month or $100/month plan. Wall Street, at least, is buying it, and is now bullish on Meta’s ability to drive revenue through Muse, Reuters reports. That’s important because the company needs to be able to justify its planned $145 billion in capital spending on compute in 2026, a figure that is up 101% from 2025.

The success of Muse took some by surprise, including Bill Nguyen, a veteran Silicon Valley entrepreneur and cofounder of Olive.is. “They caught up way faster than I thought. Like overnight,” Nguyen told Fortune. He said it looks like Meta may “win” the consumer AI race now, and there’s “no gatekeeper that prevents them from succeeding,” citing the company’s massive investments in data centers. 

Now that Meta has tasted success, Zuckerberg wants more—fast 

In Zuckerberg’s FT interview, he also reiterated his rejection of the idea of a coordinated AI slowdown across companies, or “pacing the frontier,” as it’s known within AI safety circles. That might be the best shot society has at preventing AI from inflicting serious harm, according to some industry experts. 

The most frightening risk is AI creators losing control of their models, a once abstract possibility that may be inching closer to reality. Yesterday, Anthropic published a blog post suggesting that AI models could start to self-replicate “within months.” This is most likely to occur on Internet of Things (IoT) devices, Anthropic says, listing smart ovens and TVs as examples. These devices have “no centralized point to shut off,” leaving the AIs free to reproduce themselves and spread across the web. 

But some say a coordinated slowdown could work only if all companies, in the U.S. and China, agree to it. “One defection breaks the entire strategy because everyone else has to accelerate or risk getting passed,” said one X user. “Zuck was never going to play along.”

In terms of game theory, if Meta doesn’t agree to a slowdown, it would be a case of “the weakest link”: Even if all companies besides Meta agree to a slowdown, the slowdown is only as effective as its weakest link. So, if Meta did not agree to comply, it could effectively mean no slowdown. Given the intense competition between AI companies and their perception of the battle in the industry as something of a zero-sum game, it would also be difficult for any of the AI companies to adhere to a slowdown if one of the group does not. 

Right now, Silicon Valley is split into two camps regarding AI safety risk. On the one hand, Zuckerberg and Nvidia CEO Jensen Huang promote a more accelerationist agenda, arguing that the risks have been overstated and that market forces will push AI companies to self-regulate.

On the other hand are more risk-averse leaders, most notably Anthropic CEO Dario Amodei. In September, he published a blog post calling for “speed limits” on technological development. OpenAI’s Sam Altman supported it in a rare sign of solidarity between the rival companies, following its agents autonomously hacking Hugging Face in July—as well as Australian government websites and several other entities, with activity continuing into September. 

What’s next for Muse

Meta hasn’t had a Hugging Face moment—yet. For now, its Muse assistants are actively helping people complete tasks on their computers quicker, such as responding to emails, booking flights, and surfing the web on their behalf. Some people report this is saving them loads of time, but, in exchange, they are providing Meta with more of their personal data.

Next, Meta will string the Muses together to create a team of helpful agents, Nguyen predicts. “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.” Nguyen also noted that Meta’s Ray-Ban AI glasses could be another advantage the company has to promote mainstream consumer adoption. 

To be fair, Zuckerberg isn’t positioning himself as anti-AI safety. He said Meta “delayed shipping Muse for several months to focus on safety and security,” in a September X post. Instead, he suggests the company does not need a coordinated, international slowdown to do this. “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.

It’s also unclear if “pacing the frontier” was ever a realistic option. Even though Amodei and Altman claim to support it, both continue to put out new models, particularly low-cost options that could drive mainstream adoption, while continuing to research their next boundary-pushing flagship models. Meanwhile, neither Trump nor China’s President Xi are interested in slowing development.

While the idea of slowly, safely developing a potentially society-breaking technology may sound appealing, the financial incentives may be too high for that to ever happen. Maybe Zuckerberg just said it out loud.