Meta releases Muse Wonder, first frontier AI model for complex tasks
The model is priced at US$4 per million input tokens and US$20 per million output tokens, and includes long-term active memory.
Meta launched two new AI models, Muse Wonder and Muse Wonder Full, as the company’s first frontier models for handling complex tasks, including programming. Both were developed by Superintelligence Lab and positioned above Muse Spark, the lighter model previously designed for low-latency responses. In Meta internal testing, Muse Wonder performed on par with Anthropic’s Fable 5.1 and OpenAI’s GPT-6 Astra. Meta also equipped it with active memory, a feature that lets the model retain and combine user information over days to months. Muse Wonder is available through Meta Model API, while Muse Wonder Full is limited to Muse Code customers. It is Meta’s first high-end AI model since Alexandr Wang overhauled the company’s AI development in August 2025.