Meta launches Muse Wonder and Muse Wonder Full artificial intelligence models

Meta Platforms released two artificial intelligence models named Muse Wonder and Muse Wonder Full on Monday, SiliconANGLE and TechCrunch reported. Meta told SiliconANGLE that the models are its most capable systems yet for running personal and professional agents that handle longer, multistep workflows. Wired and The Verge both reported that the models were developed by Meta's revamped Superintelligence Lab division. [1][2][3][4]

The new systems are larger and achieve higher benchmark scores on complex tasks such as coding than Meta's earlier lightweight model, Muse Spark, TechCrunch reported. SiliconANGLE reported that coding is a specific strength for the new models, alongside specialist domains including law, finance, and health. SiliconANGLE added that Meta highlighted working through an insurance claim as an example of a multistep task suited for the systems. [1][2]

According to The Verge, Meta claimed its internal testing shows the new models can rival leading systems from OpenAI and Anthropic, including GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1. SiliconANGLE reported that the models introduce a feature called Active Memory, which carries context across long-running tasks. [1][4]

The models will be accessible through an application programming interface and via Muse Code, a tool for software development, Wired reported. Wired and The Verge both reported that the models will also power features across Meta's platforms, including its personal agent named Muse. [3][4]