Zuckerberg Says Meta Has Caught Up With OpenAI and Anthropic, Promises to Move Faster

Mark Zuckerberg announced that Meta Platforms is no longer playing catch-up with its rivals at the AI frontier. The occasion was the launch of two models, Muse Wonder and Muse Wonder Full, unveiled the day before. By the company's own assessment, both match the frontier systems from OpenAI and Anthropic. Zuckerberg added that Meta (designated extremist and banned in Russia) plans to move faster from here. That line matters more than the release itself: it sets the tone for the next few quarters.

What exactly is being claimed

The models come in two versions: Muse Wonder and the larger Muse Wonder Full. The source material includes no public details on architecture, size or pricing. What it does include is a claim of parity with the flagships of Meta's main rivals, along with a reference to results on some benchmarks. The words "some benchmarks" are the key here. Parity on individual benchmarks does not mean a model has become equal across all tasks, from long agentic scenarios to writing code.

Caution is all the more warranted because the competition is not slowing down. OpenAI and Anthropic ship updates on a tight schedule, so a comparison made today may be out of date by the time the models reach mass adoption.

A bet on agents and calm about risks

The second part of Zuckerberg's statement concerns strategy. According to him, personal AI agents are becoming a priority for Meta. The logic is clear. The company has a huge audience in messengers and social networks, and an agent that acts on a user's behalf fits naturally onto that infrastructure. The open question is how reliably such an agent can handle multistep tasks, and only real-world use will answer it.

Zuckerberg also touched on safety. Meta does not see a need to slow development by force, and "machine uprising" scenarios do not worry him much. This is the position of the camp that favors speed and openness to experimentation. Some researchers take the opposite view, and they have arguments of their own. But for the market, the signal itself is what matters: Meta does not intend to take a pause.

What is known so far

• New models: Muse Wonder and Muse Wonder Full.
• Claimed level: no lower than the frontier models from OpenAI and Anthropic.
• Evidence: results on some benchmarks; the full set of metrics has not yet been disclosed.
• Strategic priority: personal AI agents.
• Pace: Zuckerberg promises to accelerate and sees no point in artificial slowdown.
• Risks: he calls scenarios of losing control over AI not particularly alarming.

What to watch next

For developers and companies choosing a model provider, the main takeaway is simple: the list of serious players may have grown longer. But the decision should rest on your own testing, not on a CEO's statement.

The practical sequence looks like this. First, wait for independent benchmarks and analyses from third-party labs. Then run the models on your own tasks: code, document processing, agentic chains with tool calls. Compare cost per request and latency separately, because in production workloads they often matter more than a gap in benchmark scores. Finally, check access terms and licenses, since for Meta this question has historically been one of the most important.

If the claimed parity holds up, OpenAI and Anthropic will have one more competitor with enormous distribution. If it does not, the gap will show on the first real tasks. Either way, the coming months will show what stands behind the talk of acceleration.