Meta has stopped playing catch-up on paper and started doing it in public. On Monday, the company rolled out Muse Wonder and Muse Wonder Full, its newest and most capable AI models, and said flatly that they stand shoulder to shoulder with OpenAI’s ChatGPT and Anthropic’s Claude.

Wall Street liked the sound of it. Meta’s shares jumped about 4 percent in afternoon trading, according to Reuters.

“Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier,” chief executive Mark Zuckerberg wrote on X, as reported by AFP.

If that reads like a victory lap, Meta can afford one. Fourteen months ago, the company tore down its AI operation and started over; the blog post announcing Wonder described that rebuild in blunt terms, saying Meta had gone from fixing its foundations to pushing the frontier. Alexandr Wang, the Scale AI co-founder Zuckerberg poached to run the effort as chief AI officer, called Wonder the company’s “largest and most capable model yet” and said it “rivals the big boys,” a jab clearly aimed at OpenAI and Anthropic.

What the models actually do

Forget chatbots that answer one question at a time. Meta pitched Wonder as an engine for long-running work: handling complex coding projects, using a computer on your behalf, working through specialized material in law, finance, and health, and, when the going gets tough, recovering from errors instead of giving up.

Two features carry most of that promise. Active Memory lets the AI remember important details about a user across multiple conversations and pull them back when they matter. Think of it as the model keeping notes instead of starting every chat from scratch. Active Focus is meant to keep an agent on task when it runs into trouble, trying other approaches, adjusting to new information, or asking for clarification instead of going silent.

There’s also scale. Both models support a context window of one million tokens, enough to feed them very large documents or codebases in a single go.

The benchmark fight

Meta published benchmark results it said came from the independent firm Artificial Analysis. The headline numbers: on the Artificial Analysis Intelligence Index, Wonder Full scored 56, Wonder scored 52, and Meta’s earlier Muse Spark 1.3 scored 48. Meta’s version of the story is that the Wonder models beat or matched OpenAI, Anthropic, and Google systems on selected coding, memory, and professional tasks.

A few caveats, because benchmarks deserve them. Meta picked the tests. Artificial Analysis’ own site notes that Wonder ranks among the most expensive models to run. And nobody outside Meta has yet run the kind of independent, head-to-head evaluation that actually settles these fights. Take the scores as a claim, not a verdict.

How you can actually use it

Here’s the rollout plan, per Meta’s announcement. Muse Wonder is available now to developers through the Meta Model API and through the company’s Muse Code programming tool. The beefier Wonder Full goes to paid Muse Code subscribers, with plans starting at $50 a month. Eventually, Wonder will also power Meta’s Muse personal agent app, but only for subscribers at first.

That paywall matters. Meta has spent years handing out AI for free through Facebook, Instagram, and WhatsApp. Wonder is the company trying to turn its billions in AI spending into subscription revenue, which is exactly why analysts were watching.

Why this is bigger than one launch

Step back and the Wonder launch reads like the receipt for one of the most aggressive talent-and-capital campaigns in tech history. In June 2025, Zuckerberg paid $14 billion for a stake in Scale AI and hired away Wang to run a newly formed Superintelligence Labs. Meta launched its Spark models in April 2026, a clean break from its open-source Llama era, which had fallen behind the industry. Wonder is the second act of that strategy, and the first one Meta claims is genuinely frontier-class.

The timing isn’t accidental either. Anthropic has been shipping new Claude models roughly monthly, OpenAI’s GPT-6 Astra line is the model to beat on most leaderboards, and Google’s Gemini keeps gaining on the consumer side. Meta isn’t just releasing a model. It’s arguing that the race for personal AI agents, the thing all four companies are really building toward, is still open, and that it belongs in the conversation.

Does Wonder actually beat Claude or GPT-6 Astra? We won’t know until independent tests land. But Meta finally has a horse worth checking, and that alone changes the race.

Why this matters

  • The agentic era is the new battleground. All four labs now compete on memory, persistence, and computer use, not raw chat quality. Wonder’s Active Memory and Active Focus are Meta’s answers.
  • Subscribers, not advertisers. Wonder debuting behind a $50-a-month paywall signals Meta is serious about AI subscriptions as a revenue line next to ads.
  • Open source is officially the past. Llama is gone as the flagship. Every major US lab now builds closed frontier models first.

Frequently asked questions

What are Meta’s Muse Wonder AI models? Muse Wonder and Muse Wonder Full are Meta’s newest and most capable AI models, launched October 12, 2026, by Meta Superintelligence Labs. Meta claims they perform on par with OpenAI’s ChatGPT and Anthropic’s Claude, and they are designed for long-running agentic tasks like coding, law, finance, and health work.

How much does Muse Wonder Full cost? The full model is available through paid Muse Code subscriptions starting at $50 per month. The smaller Muse Wonder is available to developers through the Meta Model API and the Muse Code programming tool, and will later come to Meta’s Muse personal agent app for subscribers.

What is Active Memory in Muse Wonder? Active Memory is a feature Meta built into the Wonder models that lets the AI retain important information about a user across multiple conversations and pull it back up when relevant. A companion feature, Active Focus, is meant to keep an AI agent on track when it hits problems by trying alternative approaches or asking for clarification.

Do Meta’s Muse Wonder benchmarks prove it beats ChatGPT and Claude? Not definitively. Meta published benchmark data from the independent firm Artificial Analysis showing Wonder Full scoring 56 and Wonder 52 on the Artificial Analysis Intelligence Index, ahead of Anthropic’s and OpenAI’s models on selected coding, memory, and professional tasks. But the tests were selected by Meta, and independent testing is still needed.

Sources: Reuters, AFP, Anadolu Agency, WCCFTech, MartechAI.