How will Meta's restricted Muse Wonder Full impact enterprise costs?

Meta has released two new artificial intelligence models, Muse Wonder and Muse Wonder Full, marking its first entry into the frontier AI category according to TechCrunch. These releases follow the company's restructuring of its AI research efforts last year, which included hiring former Scale AI CEO Alexandr Wang to lead the newly formed Meta Superintelligence Labs. The models represent a significant shift from Meta's previous lightweight offerings, aiming to compete directly with the most capable systems from OpenAI and Anthropic.

How do the new models compare to industry leaders?

According to Wired, Meta claims the Muse Wonder Full model possesses frontier capabilities, placing it among the world’s most capable AI systems. Testing presented by Meta shows that the models achieve benchmark scores comparable to Anthropic’s Claude Fable 5.1 and OpenAI’s GPT-6 Astra. On the Artificial Analysis Intelligence Index, a widely used aggregated score, the largest Muse version outperforms Claude Fable 5.1 but ranks just below the top-tier Claude Opus 5.5. Meta also states that the model surpasses GPT-6 Astra and both Claude versions on specialized benchmarks for legal documents and software development tasks.

What are the access and pricing implications for organizations?

Access to these models varies by version and distribution channel. TechCrunch reports that Muse Wonder Full is exclusively available to subscribers of Muse Code, a tool for writing software with AI, and will not be served through the Meta Model API. The standard Muse Wonder model is available via the API, with pricing set at $4.00 per million input tokens and $20.00 per million output tokens. This cost structure makes it roughly equivalent to GPT-6 Sol or Claude Opus 5.5, suggesting that enterprises can expect similar operational expenditures when integrating these models into their workflows compared to existing market leaders.

How do these models handle long-term user context?

Meta emphasized the models’ ability to retain and synthesize information over extended periods, a feature the company terms “active memory.” According to TechCrunch, this capability was tested using the MemoryBench benchmark, which evaluates continual learning and memory construction frameworks. Meta’s internal testing indicates that Muse Wonder scores higher on this specific metric than Anthropic’s Fable and Opus models, as well as OpenAI’s Astra. Research scientist Kavya Kopparapu stated that significant effort was dedicated to ensuring the model remembers important user details over days, weeks, and months, which is critical for personal assistant applications.

What does this mean for enterprise AI integration strategies?

The release of Muse Wonder and Muse Wonder Full suggests that Meta is aggressively targeting the high-end AI market after falling behind competitors in recent years. By offering a model that rivals top systems in coding and legal tasks, Meta provides enterprises with a new option for complex workloads. However, the restriction of the most powerful version to Muse Code subscribers may limit its immediate utility for organizations relying solely on API integrations. That suggests businesses should evaluate whether the specialized capabilities of the API-available Muse Wonder model meet their specific needs before committing to a full platform migration.

What to do and how to stay safe: Meta

• Evaluate current AI workflows to determine if the specific strengths of Muse Wonder in legal and coding tasks offer a tangible advantage over existing tools.
• Review the pricing structure of $4.00 per million input tokens against current vendor contracts to assess potential cost savings or increases.
• Test the "active memory" features of the new models in non-production environments to verify their ability to retain context over extended periods.
• Monitor the availability of larger open-weight models, as Meta has stated plans to release them in the future for on-premise deployment.

Step-by-step guide: How AI Risk Management Works: The Hidden Failure Points

General security guidance from the Sandbox Dispatch newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Is Muse Wonder Full available through the Meta Model API?

No, according to TechCrunch, Muse Wonder Full is only available to Muse Code subscribers and will not be served through the API.

How does Muse Wonder Full rank against Anthropic's models?

Wired reports that Meta says the model outperforms Claude Fable 5.1 but ranks just below Claude Opus 5.5 on the Artificial Analysis Intelligence Index.

What is the cost of using the Muse Wonder model via API?

TechCrunch states that Muse Wonder costs $4.00 per million input tokens and $20.00 per million output tokens through the Meta Model API.