Meta has introduced Muse Wonder, a new family of frontier AI models the company says delivers state-of-the-art performance on personal and professional agentic tasks. In an announcement published October 12 on its research blog, Meta unveiled two models — Muse Wonder and Muse Wonder Full — describing them as its most capable systems to date and "a major leap toward personal superintelligence."
The launch marks Meta's most direct challenge yet to OpenAI and Anthropic in the fast-growing market for AI agents, according to Axios, which framed the release as the company's bid to compete with rival frontier labs. Meta shares rose following the announcement, Quartz reported, as investors weighed whether the company's rebuilt AI organization can convert research momentum into revenue. For more context on this story, see our ongoing breaking AI news.
Two Models, Available Now
Muse Wonder and Muse Wonder Full are available today through the Meta Model API and Muse Code, the company's coding tool, according to the announcement. Muse Wonder Full, the more capable tier, is being offered to Muse Code subscribers. Meta says Muse Wonder will soon power experiences in Muse, the company's consumer assistant, with the rollout reaching subscribers first, and that the model will also handle features like artifact generation that "require more intelligence."
Pricing details were not included in the research post itself, but early coverage flagged aggressive positioning. Technology.org reported a price of $4 per million tokens, while The Neuron described the flagship as "priced like Opus," a nod to Anthropic's premium Claude tier. Artificial Analysis, the independent evaluation service, currently lists no public API pricing for the top configuration, suggesting access remains limited as the rollout begins.
Active Memory and Active Focus
The series is built around two capabilities Meta argues are essential for agents that act on a person's behalf. The first, Active Memory, carries important context about a user across conversations and actively retrieves the details relevant to the task at hand. The second, Active Focus, describes sustained effort toward a challenging goal: trying alternative approaches, working through blockers, adapting to new information, asking for clarification when needed, and checking in before taking important actions.
Meta says it evaluates these abilities with benchmarks including FocusBench, which measures focus, and SWE-Marathon, which tests sustained work on long-running software projects. Domain expertise is assessed through tasks in law, finance, and health, including Harvey's Legal Agent Benchmark (LAB), which requires models to work from source documents and follow detailed instructions to produce legal work products.
Independent Rankings Place It Near the Top
The clearest early signal comes from Artificial Analysis, whose leaderboard places Muse Wonder Full (Max) fourth out of 689 models on its Intelligence Index, with a score of 56 against a median of 12 for the field. The service classifies the model as proprietary, released in October 2026, with support for text, image, speech, and video inputs, a 1 million token context window, and reasoning enabled.
One caveat from the same data: the model appears verbose. Artificial Analysis notes it generated roughly 140 million output tokens while being evaluated, well above the 88 million median — a trait that can translate into higher real-world costs even at low per-token prices.
Trained at Prometheus
Both models were trained at Meta's gigawatt-scale Prometheus data center, according to the announcement, which frames the release as the product of a ground-up overhaul of the company's AI efforts that began fourteen months ago. "In that time, we've gone from rebuilding our stack from first principles to pushing the frontier," the company wrote. The models build on the "predictable, rigorous scaling" approach Meta outlined with Muse Spark, an earlier release, and represent what the company calls the next step on its scaling ladder.
A Crowded Week at the Frontier
The launch lands in a bruising stretch of frontier competition. Google fully released Gemini 4 Argon this week with pricing aimed squarely at OpenAI's GPT-6 Astra and Anthropic's Claude lineup, while OpenAI and Anthropic have spent recent days courting both enterprise customers and regulators. Against that backdrop, Meta is pitching persistence — agents that remember a user's context and keep working toward long-horizon goals — rather than one-shot conversation.
Whether that positioning converts into developer adoption is the open question. Meta's earlier models generally trailed OpenAI and Anthropic systems on independent leaderboards, which makes the top-five ranking for Muse Wonder Full a notable marker for an organization that, by its own account, rebuilt its AI stack from first principles barely a year ago.
Safety at the Frontier
Meta describes Muse Wonder as its "most aligned model to date," and says delivering personal superintelligence requires significant investment in alignment and safety. The company notes that a personal agent takes actions on a user's behalf and may handle sensitive context — messages, calendars, and private documents — a level of access that "demands a model that exercises discretion with private information." A full report with additional safety evaluations is promised soon.
Meta is framing the launch less as a product update than as a step in a longer arc. "This is a major technical step toward personal superintelligence," the company wrote, adding that it plans to use Muse Wonder to develop specialized models for its broader user base — a signal that the Muse Wonder series is a foundation for what comes next, not the end of it.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
