Meta launches its first high-end AI models, Muse Wonder and Muse Wonder Full
Jakarta (ANTARA) – Meta has released two new artificial intelligence (AI) models, Muse Wonder and Muse Wonder Full, its first frontier models with the ability to handle complex tasks.
TechCrunch reported on Monday (12 October), local time, that both models were developed by Meta’s Superintelligence Lab to deliver greater capabilities than Muse Spark, a lightweight AI model previously designed to provide low-latency responses.
“Together, they mark the next step in scaling our models and a major leap towards personal superintelligence,” Meta said in its official announcement.
Meta says Muse Wonder and Muse Wonder Full are larger, achieve higher benchmark scores and are better at handling complex tasks, including programming.
According to Meta’s test results, the two models scored comparably to Anthropic’s Fable 5.1 and OpenAI’s GPT-6 Astra.
Their enhanced capabilities are also reflected in their usage costs. Through the Meta Model API, Muse Wonder is priced at US$4 per million input tokens and US$20 per million output tokens. This puts it in a similar price range to OpenAI’s GPT-6 Sol and Anthropic’s Opus 5.5.
Muse Wonder Full, meanwhile, is available only to Muse Code customers and is not offered through an API. Both models will also support the capabilities of Muse, Meta’s AI agent.
In addition to improved performance on complex tasks, the two models can retain and consolidate information supplied by users over long periods, a feature known as active memory.
The feature is designed to help the models remember important information about users for days, weeks or even months.
To assess their ability to recall information, Meta used MemoryBench, a benchmark that tests continuous learning and memory formation.
According to the company’s internal test results, Muse Wonder achieved a higher MemoryBench score than Fable, Opus and Astra.
“We have put an enormous amount of effort into making sure the model can remember the important things about you for days, weeks, months and beyond,” said Meta researcher Kavya Kopparapu.
Muse Wonder and Muse Wonder Full are Meta’s first frontier AI models since Superintelligence Lab head Alexandr Wang overhauled the company’s AI development efforts in August 2025.
Following the restructuring, Meta introduced Muse Spark as a lightweight model and Muse as an AI agent. However, the company had yet to release a model with cutting-edge capabilities of the kind already developed by OpenAI and Anthropic and used as their flagship models.
Meta said Muse Wonder and Muse Wonder Full were trained at the Prometheus data centre in Ohio, United States. The facility began operating this year.