 |
| Meta has introduced Muse Wonder, a new family of AI models for agents (symbolic image) |
Meta's latest AI models arrive with impressive benchmarks and a 1-million-token context window, yet the consumer-facing Muse app remains unavailable across Europe
Meta has unveiled its newest artificial intelligence models, Muse Wonder and Muse Wonder Full, marking the company's most ambitious push into the frontier AI race yet. According to the official Meta announcement, these models represent "our most capable models to date" and are specifically designed for agentic tasks—AI that doesn't just answer questions but actually completes work on your behalf.
The timing couldn't be more strategic. As Decrypt reports, Meta's new flagship finally puts Mark Zuckerberg in serious competition with OpenAI and Anthropic, though the company still has ground to cover before claiming the top spot.
What Makes Muse Wonder Different
Meta's new models introduce two standout features that set them apart from typical chatbots. Active Memory allows the AI to carry relevant information across multiple conversations, remembering details that matter to you and your ongoing projects. Active Focus keeps the model working toward a goal even when tasks get complicated—trying different approaches, recovering from errors, and checking in before taking important actions.
Both models boast a context window of 1 million tokens, roughly equivalent to 750,000 words. That's a substantial leap that allows the AI to process lengthy documents, entire codebases, or extensive conversation histories without losing track of critical details.
The practical applications Meta highlights include handling insurance claims, understanding lease agreements, navigating personal finances, and even building working apps from scratch. These aren't hypothetical scenarios—they represent exactly the kind of real-world tasks that could make AI agents genuinely useful rather than just impressive demos.
Availability: Developers Get First Access
If you're a developer eager to experiment with these new models, you can start today. Muse Wonder is available immediately through the Meta Model API and Muse Code, Meta's coding agent. Muse Wonder Full, the more powerful variant, is limited to Muse Code subscribers.
For everyone else, the wait continues. Meta says Muse Wonder "will soon power experiences in Muse, and will roll out first to subscribers." Unfortunately, there's no timeline for when free users might gain access.
The Muse app itself—Meta's personal AI agent that operates in a secure virtual machine to browse the web, fill forms, and handle customer service tasks—remains geographically limited. The app launched in the United States in September and expanded to Canada just weeks later, as iPhone in Canada reported. Meta has remained silent about European availability.
The Europe Problem
Meta isn't alone in keeping its AI agents away from European users. Google recently integrated its Gemini Spark agent into the main Gemini chat with a Task button, but as the Gemini Apps Help page confirms, tasks in Gemini are unavailable in the European Economic Area, Switzerland, the United Kingdom, and Nigeria.
This pattern suggests deeper regulatory concerns rather than simple market prioritization. The European Union's strict approach to AI regulation and data privacy appears to be creating a de facto exclusion zone for advanced AI agent features.
Benchmark Battles: Meta's Claims vs. Independent Testing
Meta is understandably eager to highlight its competitive positioning. On the company's in-house Personal Superintelligence Index, Muse Wonder Full scores 59.1, ahead of Claude Opus 5.5 at 55.2 and GPT-6 Astra at 54.5. However, Meta hasn't published how this index is calculated, making independent verification impossible.
The company also points to impressive results on SWE-Marathon, a coding benchmark where Wonder Full scored 57.5 compared to Opus 5.5's 46.3 and Astra's 42.5. Meta does acknowledge that this comparison wasn't controlled and that its model ran at maximum effort.
Independent testing tells a more nuanced story. Artificial Analysis, which weights 10 different evaluations covering AI agents, coding, general knowledge, and scientific reasoning, places Claude Opus 5.5 at the top with a score of 58. Muse Wonder Full ties with Claude Sonnet 5.5 at 56, while the standard Muse Wonder that developers can actually purchase scores 52—one point below both GPT-6 Astra and Claude Fable 5.1.
The gap between Meta's self-reported numbers and independent rankings isn't unusual in the AI industry, but it does suggest that while Meta has closed the distance significantly, it hasn't quite reached the frontier.
Pricing and Privacy Considerations
Muse Wonder costs $4 per million input tokens and $20 per million output tokens through the API. That matches Claude Opus 5.5's pricing exactly and undercuts GPT-6 Astra, which charges $10 and $50 respectively.
For users concerned about privacy—particularly given Meta's complicated history with data handling—the company says it trained the models to avoid sharing private information without permission. A full safety report is promised "soon," though no specific date has been provided.
European users retain the right to object to Meta using their public Facebook and Instagram posts for AI training, a right that's particularly relevant given the regulatory environment that's keeping these AI agents out of the region entirely.
The Bottom Line
Meta has clearly made substantial progress with Muse Wonder, delivering models that compete credibly with offerings from OpenAI and Anthropic. The 1-million-token context window, combined with Active Memory and Active Focus, addresses real limitations that have frustrated users of earlier AI agents.
However, the company's cautious rollout strategy—limited to developers initially, restricted geographically, and with the most capable version locked behind a subscription—suggests Meta is still testing the waters rather than diving in fully. For European users watching from the sidelines, the question remains: when, if ever, will they get access to the AI agent revolution?
For now, the answer appears to be: not yet, and possibly not soon.