Meta launches new AI models to compete with OpenAI and Anthropic

Meta has launched two new artificial intelligence models as the technology company seeks to compete more directly with leading systems developed by OpenAI, Anthropic, and Google.

Muse Wonder and Muse Wonder Full are Meta's most capable models to date and are designed for personal and professional AI agents, the company said in an announcement on Monday.

The models can perform long-running tasks, use computers, and work in specialized areas including law, finance, and health. Meta said they can also handle complex coding projects, delegate work to smaller AI agents and recover from errors without abandoning the original goal.

One of their main features, Active Memory, allows the AI to retain important information about a user across multiple conversations and retrieve it when relevant.

Another feature, Active Focus, is intended to keep an AI agent working when it encounters difficulties by trying alternative approaches, adapting to new information, and seeking clarification when necessary.

Meta said the models could help users perform tasks ranging from understanding a lease or navigating an insurance claim to analyzing finances and building an application.

Muse Wonder is now available to developers through the Meta Model API and the company's Muse Code programming tool. The more capable Muse Wonder Full is available through paid Muse Code subscriptions, with plans starting at $50 per month for access to Meta's most advanced models.

Muse Wonder will later be incorporated into Meta's Muse personal assistant, initially for subscribers.

Meta published benchmark results that it said showed the models matching or outperforming some systems from OpenAI and Anthropic on selected coding, memory, and professional tasks. The broader performance of the models will still need to be assessed through independent testing.

The company said it trained the Wonder family to protect private information, refuse assistance with harmful activity, and strengthen defenses against cyberattacks. A more detailed safety report is expected to be released later.

The launch forms part of Meta's broader push toward what CEO Mark Zuckerberg has described as "personal superintelligence," AI systems capable of understanding users' goals, preferences and long-term needs.

Unlike some earlier Meta models released with downloadable weights, the company did not announce plans to make the Wonder models open source.