Meta releases Muse Wonder frontier AI models to handle complex tasks

Meta has released Muse Wonder and Muse Wonder Full, two new artificial intelligence models representing its first attempt at a frontier model. TechCrunch reported that Meta's Superintelligence Lab launched the pair on Monday. According to SiliconANGLE, Meta built the models to run personal and professional agents that handle longer, multistep tasks. [1][2]

The new systems follow the earlier release of Muse Spark, which TechCrunch noted was a lightweight model focused on low latency. TechCrunch and SiliconANGLE both reported that coding is a key strength of the larger Wonder models. SiliconANGLE added that Meta highlighted working through an insurance claim as an example of the multistep jobs the models can perform. [1][2]

The Verge reported that Meta claims the models rival top industry offerings from OpenAI and Anthropic, including GPT-6 Astra, Claude Opus 5.5, and Claude Fable 5.1, based on internal benchmark testing. Wired reported that the Muse Wonder models will be accessible through an application programming interface and via Muse Code, an artificial intelligence software development tool. [3][4]

Wired also reported that Meta plans to use the best version of the model to power features across its platforms, including its Muse personal agent. The Verge noted that the launch follows the company's efforts to revamp its Superintelligence Labs division. [3][4]