Meta takes on OpenAI and Anthropic with Muse Wonder; Aims for personal superintelligence

Meta, Mark Zuckerberg's social media giant, has announced its new generation of frontier AI models, Muse Wonder and Muse Wonder Full. The models are said to excel in personal and professional agentic tasks, and represent a “leap toward personal superintelligence.” Meta has positioned Muse Wonder at the forefront of AI development for the first time, according to independent benchmarks.

“Muse Wonder and Muse Wonder Full bring together the work of teams across Meta to deliver frontier capabilities for complex tasks at work and in everyday life,” Meta said in a company blog post. It also highlighted that the launch marks “another step toward bringing personal superintelligence to the billions of people who use our products.”

Must read: Meta Muse AI is learning more about you than you may realise: Here’s what it tracks

Meta’s Muse Wonder: What’s new?

Meta says the Muse Wonder series is here after a major overhaul of its AI efforts over the past 14 months. The company said the models are built on the foundation laid by Muse Spark, and are capable of tackling tasks across areas such as coding, finance, law, and healthcare, as well as everyday activities.

Meta teases two key features of the Muse Wonder series, which are Active Memory and Active Focus. These features help retain important information across conversations and retrieve relevant details when needed and to help them stay on track when handling difficult, multi-step assignments. These features are aimed at improving the performance of AI agents.

Meta’s Muse Wonder: Benchmarks and competition

Meta’s Muse Wonder Full is ranked in the top tier of AI models, outperforming competitors such as OpenAI’s GPT-Astra, Anthropic’s Fable 5 and Google’s Gemini 4, according to Artificial Analysis. However, the independent benchmarking organisation placed the model second on its Intelligence Index leaderboard, behind Claude Opus 5.5 Max and level with Claude Sonnet 5.5 Max.

Must read: Meta Muse AI comes to iPad with new work tools: Here’s what the AI agent can now do

The models excel across a suite of agentic, memory, and long-horizon coding benchmarks. In long-running coding tasks, Wonder Full scored 57.5, outperforming several prominent industry competitors. It also outperformed in long-term contextual memory evaluations, scoring 0.86/1 on internal recall tests. Both models excelled in software engineering tests, with Muse Wonder Full hitting 0.79/1 on pass@1 metrics across multi-repository tasks.  

Muse Wonder and Muse Wonder Full availability

Muse Wonder is available to developers through the Meta Model API and Muse Code, while Muse Wonder Full is currently available to Muse Code subscribers. Meta also plans to integrate Muse Wonder into its Muse AI assistant.

The company also intends to use the models to power content and application generation, as well as develop specialised models for future Muse experiences.