Meta Unveils Muse Wonder AI Models to Strengthen Its Position in the AI Race

Meta has introduced two advanced AI models, Muse Wonder and Muse Wonder Full, as it looks to strengthen its position in the global artificial intelligence race against OpenAI, Google and Anthropic. Announced on October 12, 2026, the models are designed to power personal and professional AI agents, with a particular focus on software development, financial analysis and complex legal tasks. Meta says two key features set them apart: Active Memory, which helps models retain relevant user context across conversations, and Active Focus, which allows AI agents to work through difficult tasks, adapt to obstacles and continue until they are completed. The models can process text, images, video and voice inputs, while generating text-based outputs.

According to the report, Muse Wonder has a one-million-token context window and is available to developers through Meta Model API and Muse Code, while Muse Wonder Full is restricted to Muse Code subscribers. The reported benchmark results show Wonder Full scoring 56 on the Intelligence Index, while the standard Wonder model scores 52. Meta also says the models have improved capabilities in coding and professional workflows, alongside enhanced safeguards against cybersecurity and other serious forms of misuse. However, their real-world performance, privacy implications and safety claims will need further independent evaluation. Meta says the models will also power its Muse AI app, with access initially limited to subscribers.