Meta launches Muse Wonder and Wonder Full AI models: How they compare with GPT-6

Meta has released Muse Wonder and Muse Wonder Full, two advanced AI models that mark the company’s latest attempt to compete at the frontier of artificial intelligence. Developed by Meta’s Superintelligence Lab, the models are designed for more demanding tasks, including coding and retaining information about users over time.

The launch follows the earlier release of Muse Spark, a lighter model built for faster responses. Meta says the new models deliver stronger benchmark results and perform comparably to Fable 5.1 and GPT-6 Astra in its testing.

The company describes the release as a step towards what it calls “personal superintelligence”, with a particular focus on AI assistants that can remember and use information across conversations.

Meta Muse Wonder focuses on memory and complex tasks

One of the main features highlighted by Meta is “active memory” the ability to retain and combine information shared by a user over days, weeks and months. Meta says Muse Wonder scored higher on MemoryBench, a benchmark designed to test continual learning and memory construction, than Fable, Opus and Astra in the company’s evaluations. Research scientist Kavya Kopparapu said Meta had put significant effort into helping the model remember information that matters to users over extended periods. This capability could make AI assistants more useful for ongoing tasks, such as planning projects or adapting recommendations. However, the benchmark results cited here are Meta’s own claims and do not independently establish how the models will perform in everyday use.

Muse Wonder pricing and availability explained

Meta Wonder will be available through the Meta Model API at $4 per million input tokens and $20 per million output tokens. Input tokens represent information sent to the model, while output tokens are the content it generates.

Muse Wonder Full will not be available through the API. Instead, access will be limited to Muse Code subscribers. Both models will also power Meta’s Muse agent, according to the company.

Meta steps up its frontier AI competition

The release is the first frontier-model launch since Scale AI founder Alexandr Wang restructured Meta’s AI operations in August last year, according to the supplied information. Meta says the models were trained at its Prometheus data centre in central Ohio, which came online earlier this year.

The launch puts Meta more directly into the competition with AI developers such as OpenAI and Anthropic. The key test will be whether its reported gains in memory, coding and complex reasoning translate into useful results for developers and everyday users.