Meta launches Muse Wonder models for coding and extended AI tasks

Meta introduced two AI models on October 12: Muse Wonder and Muse Wonder Full. The company says they improve its performance on software development and tasks that require an AI agent to work through multiple steps.

Muse Wonder is available through Meta's API and Muse Code. Wonder Full requires a qualifying Muse Code subscription. Meta lists a context capacity of 1 million tokens for both models.

The models build on Muse Spark and introduce features Meta calls Active Memory and Active Focus. The first retains relevant information between conversations, while the second is designed to help agents recover from mistakes and continue difficult tasks. These capabilities target work that extends beyond a single exchange.

Meta plans to bring Wonder to its Muse assistant, starting with subscribers, but has not specified a launch date. The company also says a fuller account of its safety testing will follow.