Meta, the social media giant led by Mark Zuckerberg, has announced its next generation of frontier artificial intelligence models: Muse Wonder and Muse Wonder Full. These advanced AIs represent a significant overhaul of Meta's AI strategy over the past 14 months, building upon the foundation established by Muse Spark. The company positions Muse Wonder at the forefront of AI development, with a stated goal of achieving "personal superintelligence" for billions of users.
A Leap Towards Personal Superintelligence
According to Meta, the Muse Wonder series is engineered to excel in a wide array of personal and professional agentic tasks. This includes complex operations across fields such as coding, finance, law, and healthcare, alongside everyday activities. The company highlights that this launch marks a crucial step toward integrating advanced AI capabilities into the products used by its vast global audience.
Key Features: Active Memory and Active Focus
Two standout features teased for the Muse Wonder series are Active Memory and Active Focus. Active Memory allows the AI models to retain vital information across extended conversations, ensuring relevant details are retrieved precisely when needed. Active Focus, on the other hand, helps the AI agents maintain concentration and stay on track when tackling difficult, multi-step assignments. These functionalities are critical for improving the overall performance and reliability of AI agents in complex scenarios.
Benchmarking Against Industry Leaders
In a competitive landscape, Muse Wonder Full has demonstrated strong performance against established rivals. Independent benchmarking by Artificial Analysis ranked Muse Wonder Full in the top tier of AI models, surpassing competitors like OpenAI’s GPT-Astra, Anthropic’s Fable 5, and Google’s Gemini 4. While it secured the second position on the Intelligence Index leaderboard, it achieved parity with Claude Sonnet 5.5 Max and was just behind Claude Opus 5.5 Max.
The models showcased particular strength across a suite of agentic, memory, and long-horizon coding benchmarks. Muse Wonder Full scored 57.5 in long-running coding tasks, outperforming several prominent industry competitors. It also excelled in long-term contextual memory evaluations, achieving a score of 0.86/1 on internal recall tests. Both Muse Wonder models performed exceptionally well in software engineering tests, with Muse Wonder Full hitting 0.79/1 on pass@1 metrics across multi-repository tasks.
Availability and Future Vision
Muse Wonder is currently accessible to developers via the Meta Model API and Muse Code. The more advanced Muse Wonder Full is available to Muse Code subscribers. Meta plans to integrate Muse Wonder into its Muse AI assistant, enhancing its capabilities. Furthermore, the company intends to leverage these powerful models for content and application generation, as well as to develop specialized models for future Muse experiences, pushing the boundaries of what personal AI can achieve.