Meta releases Muse Wonder and Muse Wonder Full AI models

In short: Meta has released two new AI models, Muse Wonder and Muse Wonder Full, and says they are its first “frontier” level models.

What happened

Meta’s Superintelligence Lab announced two new AI models on Monday: Muse Wonder and Muse Wonder Full. An AI model is the engine behind a chatbot or assistant, like the brain inside an app.

Meta describes these as its first “frontier model” effort, meaning a top tier system the company believes can compete with the most capable models available. Meta says the new models score well on common tests and do better on harder tasks, including writing and understanding code.

The company is selling access through the Meta Model API. An API is like a checkout counter for software, it lets other apps pay to use the model. Meta says Muse Wonder will cost $4.00 per million input tokens and $20.00 per million output tokens, where tokens are small chunks of text (roughly like pieces of words) used to measure usage.

Meta says Muse Wonder Full will not be available through the API. It will only be available to Muse Code subscribers. Meta also plans to use both models to power its Muse agent.

A key focus is what Meta calls “active memory.” Meta says the models can better hold onto and combine information a user shares over time, more like a notebook that keeps being updated instead of starting fresh each day.

Why it matters

More capable models, and especially better memory, could make AI assistants more useful for ongoing work, like longer projects, repeated tasks, or help that spans weeks or months. It also matters for cost and access, since Meta is putting the stronger “Full” version behind a separate subscription and limiting how developers can use it.

Source: TechCrunch AI