Meta Launches Muse Wonder, Its First Frontier AI Model

Meta has taken a major step forward in its artificial intelligence ambitions with the launch of two advanced AI models, Muse Wonder and Muse Wonder Full. Released on Monday by the company’s Superintelligence Lab, the new models mark Meta’s first attempt at building a frontier AI model designed to compete with leading systems from OpenAI and Anthropic.

The release signals a shift in Meta’s AI strategy, moving beyond lightweight models toward more powerful systems capable of handling complex tasks, including coding, reasoning, and long-term information management.

Muse Wonder Targets Advanced AI Performance

Meta’s earlier Muse Spark model focused on delivering fast responses with low latency. In contrast, Muse Wonder and Muse Wonder Full are larger models designed to provide stronger performance on demanding tasks.

According to benchmark results shared by Meta in the models’ documentation, both models achieve scores comparable to Fable 5.1 and GPT-6 Astra. The company also highlighted improvements in coding and other complex workloads, positioning the new releases as a significant upgrade over its previous lightweight offering.

Meta described the launch as an important milestone in its broader AI roadmap.

“Together, they mark the next step on our scaling ladder and a major leap toward personal superintelligence,” the company said in its announcement.

The release reflects Meta’s growing focus on building AI systems that can do more than answer individual questions, with an emphasis on helping users manage tasks and information over extended periods.

Muse Wonder Pricing and Availability

The new models come with different access options, depending on the version users want to use.

Muse Wonder is available through the Meta Model API, with pricing set at $4 per million input tokens and $20 per million output tokens. Based on the pricing announced by Meta, it sits in a similar cost range to GPT-6 Sol and Opus 5.5.

Muse Wonder Full, meanwhile, will be available exclusively to Muse Code subscribers. Meta will not offer this version through its API.

Both models will also power the Muse agent, extending their capabilities to Meta’s AI agent experience.

The different access arrangements suggest that Meta is targeting both developers who want to integrate AI into their applications and subscribers looking for more advanced coding and productivity tools.

Meta Puts the Spotlight on AI Memory

One of the most notable features of the new models is their ability to remember and organize information supplied by users. Meta calls this capability “active memory,” and it is central to the company’s vision of developing more personalized AI assistants.

Rather than treating every interaction as an isolated conversation, Meta wants its AI systems to retain useful details and bring them back when they become relevant. This could help an assistant provide more consistent support across tasks that unfold over days, weeks, or even months.

To demonstrate these capabilities, Meta highlighted results from MemoryBench, a benchmark designed to evaluate continual learning and memory-building frameworks.

According to the company’s testing, Muse Wonder scores higher on MemoryBench than Fable, Opus, and Astra.

“We put a tremendous amount of effort into making sure your model can remember what’s important to you over days, weeks, months, and beyond,” research scientist Kavya Kopparapu said in a video released alongside the launch.

Although benchmark results provide one way to compare AI systems, real-world performance will ultimately depend on how reliably these memory features work across different users and situations.

A New Chapter in Meta’s AI Strategy

The Muse Wonder release is also significant because it represents Meta’s first frontier-model launch since Scale AI founder Alexandr Wang restructured the company’s AI efforts in August of last year.

Since that restructuring, Meta has introduced the lightweight Muse Spark model and the Muse agent. However, the company had largely stayed away from releasing the kind of state-of-the-art models associated with competitors such as OpenAI and Anthropic.

Muse Wonder suggests that Meta is now making a stronger push into the high-performance AI market, where model quality, reasoning capabilities, coding performance, and infrastructure costs all play important roles.

The company also disclosed that the new models were trained at its Prometheus data center in central Ohio, which came online earlier this year. The facility provides part of the infrastructure supporting Meta’s broader AI development efforts.

What Muse Wonder Means for Meta’s AI Ambitions

With Muse Wonder and Muse Wonder Full, Meta is expanding its AI portfolio beyond fast, lightweight models and moving toward more capable systems built for complex tasks and personalized assistance.

The emphasis on active memory is particularly important to the company’s long-term plans. If these systems can reliably retain relevant information and use it appropriately, they could become more useful for ongoing projects, coding workflows, and everyday productivity.

However, Meta will still need to demonstrate how its models perform in practical use compared with competing AI systems. For now, the launch marks a clear change in direction: Meta is stepping further into the frontier AI race while building toward its vision of personal superintelligence.