Meta Claims AI Frontier Parity with New Wonder Models Featuring Active Memory and Focus

Meta asserts it has caught up to the frontier in artificial intelligence with the introduction of its latest Wonder AI models, Muse Wonder and Muse Wonder Full. This achievement follows over a year of intensive effort and an investment of tens of billions of dollars in rebuilding its AI capabilities, positioning these models as a significant stride towards 'personal superintelligence'.

The new models are specifically designed to excel in complex legal, financial, and health-related tasks, alongside delivering 'frontier coding performance'. A key innovation in both models is the addition of 'active memory' and 'active focus' abilities, which Meta states will enable users to accomplish more over extended periods.

Meta plans to integrate the Wonder models into its personal agent, Muse, initially for paid subscribers. 'Active memory' is defined as the capacity to retain relevant information across conversations, retrieving details pertinent to the task at hand. 'Active focus' will allow the agent to maintain progress towards a goal, even when faced with challenging tasks.

This development marks a notable milestone for Meta, which has projected spending up to $145 billion this year to fuel its ambitious AI strategies. CEO Mark Zuckerberg commented to the Financial Times that reaching this point demanded substantial computing power, investment, and rigorous scientific work from the company's superintelligence team.

The Muse Wonder and Muse Wonder Full models are currently rolling out to subscribers. Muse Wonder is accessible via Meta's API and within Muse Code, priced at $4 per million input tokens and $20 per million output tokens. Muse Wonder Full is available to subscribers of the $50/month 'power usage' or $200/month 'maximum usage' tiers of Muse Code.

What to watch: The integration of these models into Meta's products and their impact on user experience.

Editor's note: Strong coverage of Meta's new AI models, including pricing, technical capabilities, and executive commentary.