Meta has officially announced its new frontier artificial intelligence models, Muse Wonder and Wonder Full, designed to compete directly with OpenAI and Anthropic. This release represents a significant advancement over the Muse Spark model, shifting focus toward advanced reasoning, programming, and complex task execution through autonomous agents.  

According to tests presented by the company, the new models achieved performance results comparable to GPT-6 Astra and Claude Fable 5.1. Independent analyses also position the Wonder Full model among the most advanced systems on the market, particularly for programming and multi-step professional activities.  

Key technological features include: 

• Active Memory: A feature that allows the AI to store and retrieve relevant information from previous conversations and activities, even after weeks or months.  

• Active Focus: A technology designed to keep the artificial intelligence focused on its objectives, even when obstacles, errors, or unexpected situations arise during a task.  

The models were trained at the Prometheus data center in Ohio, USA, marking a new phase for Meta in developing agents capable of executing prolonged tasks with greater autonomy. The Muse Wonder model is available through the Meta Model API and Muse Code, priced at $4 per million input tokens and $20 per million output tokens, while the Wonder Full version remains exclusive to Muse Code subscribers. 

Filed in . Read more about AI (Artificial Intelligence) and Meta.