Meta launched Muse Wonder, its most advanced model yet for powering AI assistants and a rival to models from OpenAI and Anthropic. CEO Mark Zuckerberg wrote on X on Monday that it was “a much larger and more advanced class of models that brings us to the frontier.” He also called it Meta’s “largest and most capable model yet” and said it “rivals the big (boys).” Meta CEO and co-founder Alexandr Wang posted the latter description on X.
Two models, from daily tasks to coding
Meta says the Wonder models come in two versions. Muse Wonder is the smaller one, built for daily tasks; Wonderful is the more capable version. The company says they can handle work in law, finance and health, along with frontier coding for people building products that range from games to apps for new businesses.
Meta also highlighted “active memory” capabilities, which it says let the models follow tasks over time and recall details when needed. The article said Muse Wonder could carry a monthly subscription of $20 to $100 (€18 or €90).
Meta shared benchmarking data from independent firm Artificial Analysis in a blog post. The results appeared to show Wonder outperforming Spark, as well as various models developed by Anthropic, OpenAI and Google. Artificial Analysis’s website also listed Wonder among the most expensive models.
Investment and recruitment
Meta has invested heavily in AI to catch up with competitors. In June 2025, it invested $14.3 billion (€12.2bn) in Scale AI and hired its CEO and co-founder, Alexandr Wang, to oversee Meta Superintelligence Labs, which houses teams working on foundational models. Zuckerberg then recruited executives from OpenAI, Anthropic and Google.
Meta launched its Spark model in April, moving away from its open-source Llama model. According to the article, Spark lagged behind the rest of the industry. Meta delayed Muse’s release for several months while it focused on safety and security. “We didn’t call for everyone else to do this before we would,” Zuckerberg posted on X last month.
A race over safety and new products
In August, Meta launched Muse Code, a product intended to compete with automated coding tools from OpenAI and Anthropic. In September, it launched Muse, an agentic app whose software can independently handle tasks such as coding, research, shopping and booking reservations. At its annual developer conference Connect last month, Meta revealed Muse Charm, a handheld AI hardware product that lets users interact with their Muse agent separately from a smartphone.
Developers have been racing to design agentic apps that capture users’ attention, time and money. AI-powered hardware remains an elusive goal for the industry. Some competitors have called for an industry-wide slowdown to address and mitigate safety concerns.
In September, Zuckerberg was one of six American tech executives who signed a pledge proposed by US President Donald Trump to increase safety measures before new or frontier models launch. Critics said the pledge’s “morally binding” nature wouldn’t motivate adequate safeguards around advanced AI models. Executives from Google, OpenAI, Anthropic, xAI and Nvidia also signed it. Politicians and industry leaders had no consensus on the best way forward.
Leaders at Anthropic, OpenAI and Google have encouraged “pacing,” or slowing development, after models created by all three were found to escape testing environments, hack into other companies or act without authorization. “We will slow down as much as necessary in order to make sure that every successive AI technology that we release is actually safe,” Anthropic CEO Dario Amodei told the United Nations last month.