Meta Launches First Frontier AI Model, Muse Wonder

A New Entrant in the Frontier Race

Meta has released its first frontier-class AI model, called Muse Wonder, along with a larger version named Muse Wonder Full. The announcement came from Meta's Superintelligence Lab on Monday and marks the company's first serious push into the same high-end AI space where OpenAI and Anthropic dominate.

Where Meta's earlier Muse Spark model was built to be small and fast, the new Muse Wonder models are bigger and more powerful. They score near the top on benchmark tests — matching or beating systems like GPT-6 Astra and Fable 5.1 Artificial Analysis.

Built for Real Work

Meta is positioning these models as tools that can handle complicated, real-world tasks. The company says they perform well in areas like law, finance, and health, as well as in coding — helping people build apps and games.

Two features stand out. The first is what Meta calls "Active Memory" — the ability to remember important information across long conversations over days, weeks, or even months. Meta tested this on its own MemoryBench benchmark and says Muse Wonder scores higher than Claude Opus 5.5, Fable, or GPT-6 Astra.

The second is "Active Focus" — the ability to keep working toward a goal even when things get difficult, recover from mistakes, and ask for help when needed.

Pricing and Access

Muse Wonder is available through the Meta Model API at a price point similar to other top models: $4 per million input tokens and $20 per million output tokens. This puts it roughly on par with GPT-6 Sol and Opus 5.5.

Muse Wonder Full, the larger version, will only be available to subscribers of Muse Code, Meta's coding-focused plan. It will not be accessible through the API.

Both models will also power Muse, Meta's personal AI agent, with Wonder starting to roll out to subscribers first.

Behind the Scenes

The models were trained at Meta's Prometheus data center in central Ohio, which came online earlier this year. According to Meta, these releases come after a major restructuring of the company's AI efforts led by Alexandr Wang, who took over in August of last year. Since then, Meta had focused on lighter models like Muse Spark while competitors pushed forward with frontier systems.

Meta says it follows safety guidelines outlined in its Meta Superintelligence Scaling Framework and that a full safety report will be released soon.

Key facts

• Muse Wonder is Meta's first frontier-class AI model, matching top competitors on benchmarks
• Active Memory lets the model retain important user information across long periods
• Muse Wonder Full is restricted to Muse Code subscribers and not available via API
• Pricing is $4 per million input tokens and $20 per million output tokens
• Models were trained at Meta's Prometheus data center in Ohio

Sources

1. Meta releases its first frontier model, Muse Wonder | TechCrunch — techcrunch.com
2. Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents — about.fb.com
3. Muse Wonder: A Frontier Model For Personal & Professional AI Agents
4. Muse Wonder | Meta
5. Muse Wonder and Muse Wonder Full: Meta's New Frontier AI Models — intelligentliving.co