Meta has introduced two new frontier-level AI models, Muse Wonder and Muse Wonder Full, designed to enhance its personal agent and coding tools with improved memory and task-handling capabilities.
Key facts
- •Meta released two frontier-level models, Muse Wonder and Muse Wonder Full, on Monday.
- •The models feature 'active memory' to retain user information and 'active focus' to persist through difficult tasks.
- •Muse Wonder is available via the Meta Model API at $4 per million input tokens and $20 per million output tokens.
- •Muse Wonder Full is exclusive to Muse Code subscribers on higher-tier plans.
- •The models were trained at Meta’s Prometheus data center in central Ohio.
Meta has launched two new artificial intelligence models, Muse Wonder and Muse Wonder Full, marking the company’s first foray into frontier-level AI development. The models are designed to power Meta’s personal agent and coding applications, featuring enhanced capabilities for complex tasks and long-term memory retention. These releases follow a significant restructuring of Meta’s AI research division led by Alexandr Wang and represent a major investment in the company's long-term AI strategy.
By the numbers
Capabilities and Features
The new models introduce features Meta calls "active memory" and "active focus," intended to help AI agents retain information across long periods and persist through complex tasks without stalling. According to Meta, the models are particularly effective at coding, as well as handling legal, financial, and health-related tasks. Testing presented by the company indicates that the models achieve benchmark scores comparable to or exceeding those of leading competitors like Fable 5.1, Opus 5.5, and GPT-6 Astra.
Availability and Pricing
Muse Wonder is accessible through the Meta Model API, priced at $4 per million input tokens and $20 per million output tokens. It is also available to users of the Muse Code tool. Muse Wonder Full, the more capable version, is restricted to Muse Code subscribers on specific "power usage" or "maximum usage" tiers. Meta noted that the models were trained at its Prometheus data center in central Ohio.
Strategic Context
The release follows a period of intense investment in Meta’s Superintelligence Labs, with the company projecting AI-related spending to reach as much as $145 billion this year. Meta AI chief Alexandr Wang stated that the models were developed to improve the reliability of personal and professional agents, specifically by training them to navigate roadblocks in real-world tasks. The company also noted that the new models include its strongest safeguards to date against cyber misuse.
Advertisement
This article was independently rewritten by ManyPress editorial AI from reporting originally published by TechCrunch, CNBC Technology, Wired, Engadget.
