Four dollars for one million input tokens, and a promise to match: Mark Zuckerberg calls his newest model “our biggest and most capable model yet.” With Meta Muse Wonder and the more powerful variant Muse Wonder Full, Meta released two AI models on October 12 and 13, 2026 that, according to independent benchmarks, can keep pace with the flagship models from Anthropic and OpenAI. For the global tech community, this is a significant moment: Meta is once again positioning itself as a serious contender in the race for the most advanced AI technology.
The launch comes as no surprise, but at a strategically smart moment. After months of preparation and, according to Zuckerberg’s own statements on X, additional safety work, Meta is now bringing two models to market aimed at both developers and enterprises in law, finance, and healthcare.
Table of Contents
Key Takeaways
- Meta released two new AI models on October 12/13, 2026: Meta Muse Wonder and Muse Wonder Full.
- According to Artificial Analysis benchmarks, Muse Wonder Full outperforms Anthropic’s Claude Fable 5.1 and comes in just below Claude Opus 5.5.
- The models bring two new features: Active Memory and Active Focus.
- Muse Wonder is available via the Meta Model API starting at $4 per million input tokens; Muse Wonder Full is exclusive to Muse Code subscriptions.
- The launch follows Meta’s billion-dollar investment in Scale AI and the founding of Meta Superintelligence Labs under Alexandr Wang.
What exactly is Meta Muse Wonder
Meta Muse Wonder is the smaller of the two new models and was designed for everyday tasks. It targets developers and businesses that need a capable yet cost-efficient AI solution. The larger variant, Muse Wonder Full, is currently Meta’s most powerful model and is accessible exclusively through paid Muse Code subscriptions.
This two-tier approach is no coincidence. Meta is following a pattern also seen among competitors: a broadly accessible base model for mass use, complemented by a premium model for demanding, compute-intensive tasks. Anyone who wants to use Muse Wonder Full needs a Power or the new Maximum usage subscription from Muse Code.
Mark Zuckerberg’s announcement
Zuckerberg himself chose strong words for the announcement. On the platform X, he described Muse Wonder Full as “our biggest and most capable model yet,” one that “rivals the big players” — a clear nod to the industry’s established heavyweights. This phrasing is notable because Meta is openly inviting a direct comparison with Anthropic and OpenAI rather than limiting itself to its own niche applications.
Alexandr Wang, CEO of Meta Superintelligence Labs since the Scale AI acquisition in June 2025, also commented on the launch. He described the new models as a step toward the so-called AI frontier, meaning the cutting edge of technological development. Wang’s assessment carries weight, since before moving to Meta he was deeply involved in developing and evaluating AI systems as co-founder of Scale AI.
“Our biggest and most capable model yet, one that rivals the big players.” — Mark Zuckerberg on X
Pricing and availability at a glance
For developers and businesses, pricing is likely of particular interest. Meta Muse Wonder is available immediately via the Meta Model API, with the following pricing structure:
| Model | Access | Price (per million tokens) |
|---|---|---|
| Muse Wonder | Meta Model API | $4 input / $20 output |
| Muse Wonder Full | Muse Code (Power or Maximum plan) | Subscription only |
This pricing structure shows that Meta is deliberately targeting two audiences: developers who access Muse Wonder directly via the API and pay per token, and power users and businesses who get access to the more capable Muse Wonder Full through a subscription.
How does Meta Muse Wonder compare
The key question for the industry is whether Meta Muse Wonder can actually keep up with the established models. According to Meta itself and independent benchmarks from Artificial Analysis, the answer is nuanced.
In the Artificial Analysis Intelligence Index, one of the most closely watched independent benchmarks for large language models, Muse Wonder Full outperforms Anthropic’s Claude Fable 5.1. At the same time, it lands just below Claude Opus 5.5, also an Anthropic model. In other words: with Muse Wonder Full, Meta has caught up to the upper-middle tier of the AI industry without reaching the absolute top spot.
Things get interesting with specific use cases. Meta states that Muse Wonder Full outperforms OpenAI’s GPT-6 Astra in certain benchmark categories, including legal document processing and coding tasks. This claim comes from Meta itself and refers to selected test categories, not a general superiority across all use cases.
Putting the benchmark results in context
| Comparison | Result according to Meta/Artificial Analysis |
|---|---|
| Muse Wonder Full vs. Claude Fable 5.1 (Anthropic) | Muse Wonder Full outperforms it |
| Muse Wonder Full vs. Claude Opus 5.5 (Anthropic) | Slightly behind |
| Muse Wonder Full vs. GPT-6 Astra (OpenAI) | Better in specific areas like law and coding |
It’s important for readers to keep the attribution clear: Claude Fable 5.1 and Claude Opus 5.5 come from Anthropic, GPT-6 Astra from OpenAI. Meta Muse Wonder and Muse Wonder Full are exclusively Meta’s own developments and compete directly against these outside models — they are not joint projects or licensed products.
New features: Active Memory and Active Focus
Beyond raw performance, Meta Muse Wonder brings two features designed to improve everyday practical use.
Active Memory allows the model to remember important details about a user over a longer period of time. In practice, this means returning users don’t have to re-enter certain context information, since the system retrieves relevant details from earlier interactions.
Active Focus, on the other hand, targets complex, long-running tasks. This feature is designed to keep the model working consistently on a task even through obstacles or interruptions, instead of losing track. This can make a noticeable difference especially in multi-step coding projects or complex analysis tasks.
Both features address a well-known problem with large language models: losing context over longer interactions or complex workflows. With Active Memory and Active Focus, Meta is positioning its models specifically for professional, demanding use cases.
Use cases: from law to software development
According to Meta’s own statements, the Muse Wonder models are designed for four central application areas:
- Law: processing and analysis of legal documents
- Finance: support for finance-related tasks and analysis
- Healthcare: use in documentation-intensive healthcare processes
- Coding: from small apps to complex software development
This breadth of use cases shows that Meta Muse Wonder is not primarily intended as a consumer chatbot, but as a professional tool for industries with high demands on precision and reliability. The coding segment in particular should be of interest to developers worldwide, since this is where direct competition with existing tools like GitHub Copilot and other AI-assisted development environments is especially pronounced.
The road to Meta Muse Wonder: strategic background
The launch of Meta Muse Wonder doesn’t stand in isolation — it’s part of a longer strategic development at Meta. Back in August 2026, the company released the coding tool Muse Code, followed by the agentic app Muse in September 2026. Muse Wonder and Muse Wonder Full thus mark the current high point of a series of product releases built on top of each other within just a few months.
It’s also notable that, according to his own statement on X, Zuckerberg deliberately delayed the release of the Muse models to carry out additional safety work. This suggests that Meta paid particular attention to model safety ahead of the broad rollout, before making the models available to paying customers and developers.
The role of Alexandr Wang and Meta Superintelligence Labs
A central building block for this progress was Meta’s investment in Scale AI. In June 2025, Meta invested €1.43 billion, roughly $14.3 billion, in Scale AI and brought on its co-founder Alexandr Wang as head of the newly created Meta Superintelligence Labs. This investment is considered one of the largest single investments in the AI sector in recent years and marked a clear strategic shift at Meta, away from a pure social media focus and toward a serious position in the race for advanced AI systems.
The current release of Muse Wonder and Muse Wonder Full can thus be interpreted as the first major visible result of this strategic realignment under Wang’s leadership.
Competitive landscape: Anthropic and OpenAI compared
To fully grasp the significance of Meta Muse Wonder, it’s worth looking at the competitive landscape so far. Anthropic has two models in its portfolio, Claude Fable 5.1 and Claude Opus 5.5, which have so far ranked among the most capable on the market. Claude Opus 5.5 in particular is, according to the benchmarks mentioned, still considered the stronger of the two Anthropic models and remains just ahead of Meta’s new flagship model even after the Muse Wonder launch.
OpenAI, meanwhile, is represented in the market by GPT-6 Astra — see also our coverage of GPT-6’s self-building interface features. According to Meta’s own statements, Muse Wonder Full performs better than GPT-6 Astra in certain specialized benchmark categories, such as legal documents and coding tasks. However, this does not automatically mean general superiority across all use cases, but rather points to targeted strengths in specific usage scenarios.
Overall, the market for large AI models remains a three-way race between Meta, Anthropic, and OpenAI, with no single company leading across the board in every category. Each company has specific strengths: Anthropic with Claude Opus 5.5 still at the top of the Artificial Analysis Intelligence Index, Meta with targeted advantages in law and coding, and OpenAI as the established third major player with GPT-6 Astra.
Conclusion and next steps
Meta Muse Wonder and Muse Wonder Full mark a significant step in Meta’s AI strategy. With competitive benchmark results, new features like Active Memory and Active Focus, and a clear pricing structure, Meta is positioning itself as a serious challenger to Anthropic and OpenAI. The combination of an affordable entry-level model via the API and an exclusive premium model via Muse Code in particular shows a well-considered market strategy.
For developers and businesses evaluating AI solutions for law, finance, healthcare, or software development, it’s worth taking a close look at the pricing structure and the specific benchmark strengths of each model. Anyone already working with Muse Code can get direct access to Muse Wonder Full through a Power or Maximum subscription. Those looking for a cost-efficient entry via the API will find Muse Wonder a solid alternative for everyday use cases. The coming months will show how these models hold up in practical use against Claude Opus 5.5 and GPT-6 Astra.
