Meta Unveils Frontier ‘Wonder’ AI Models to Challenge OpenAI, Anthropic

Meta Platforms Inc. unveiled its most powerful artificial intelligence (AI) models to date on Monday in a stepped-up challenge to OpenAI and Anthropic in the escalating race to commercialize next-generation AI software.

The new frontier models, named Wonder and Wonder Full, mark a major advancement in Meta CEO Mark Zuckerberg’s push to pivot the social media giant beyond its core advertising business. Speaking on X, Zuckerberg characterized the releases as a leap into an advanced class of technology following the debut of Meta’s earlier Spark model family in April.

According to benchmark data from independent research firm Artificial Analysis, the Wonder models outperform Meta’s previous offerings as well as comparable systems from OpenAI, Anthropic, and Alphabet’s Google. However, the data also indicates that Wonder ranks among the most expensive models currently on the market.

The launch highlights a multibillion-dollar effort by Meta to reshape its standing in Silicon Valley.

In July, the Facebook and Instagram parent company disclosed that its AI-related capital spending could reach $145 billion this year. The aggressive strategy included a $14 billion investment in San Francisco-based startup Scale AI in June 2025, alongside the high-profile recruitment of Scale AI co-founder Alexandr Wang. Wang now serves as Meta’s Chief AI Officer, heading up its newly established Superintelligence Labs division.

The new models will soon power Meta’s subscription-based personal agent app, Muse, which launched in September. Designed to execute tasks autonomously such as research, coding, and booking reservations, Muse represents Meta’s bid to capture consumer attention and subscription revenue across its global base of 3.6 billion daily active users.

Meta said Wonder Full will also be made available to subscribers of its automated coding assistant, Muse Code. At its annual Connect conference last month, the company further expanded its hardware footprint by previewing Muse Charm, a standalone handheld AI device.

HyperFRAME Research analyst Stephanie Walter said Meta wants Wonder to “remember your context and keep working until the job is done.”

“That could make personal assistants and coding agents more useful, but persistence can also mean spending longer pursuing the wrong answer,” Walter said. “The test is whether these models complete tasks correctly and recognize when they need help.

“Meta has published safety assessments, so there is evidence to examine. But it says more evaluations are coming. I want to know what remains outstanding and what it means for trusting these agents with private information and authority to act.”

Meta’s stock rose approximately 4% in Monday afternoon trading following the announcement.

The rapid rollout comes amid broader debates over AI safety and oversight. Unlike Meta’s earlier open-source Llama models, the company delayed the release of its Muse suite by several months to conduct internal safety checks. Zuckerberg noted last month that Meta sought to balance development speed with security measures.

The industry remains divided on the pace of deployment. In September, Zuckerberg was among six prominent tech executives to sign a voluntary commitment backed by President Donald Trump to implement safeguards around advanced models.

Meanwhile, rival executives — including Anthropic CEO Dario Amodei — have advocated for “pacing” after several top-tier models previously acted without authorization during internal testing.

Despite caution from competitors and regulatory scrutiny, tech companies continue to accelerate product cycles. Industry reports indicate Zuckerberg expedited the launch of Muse in part due to the rapid growth of viral AI startups like Instinct, signaling that the battle for dominance in consumer AI remains fierce.