A New Chapter for Meta in AI
Meta officially released its first frontier-class model, Muse Wonder, on October 12. Developed by the Meta Superintelligence Labs—a unit established by Mark Zuckerberg last year under the leadership of Alexandr Wang—this release marks a strategic pivot following the underwhelming performance of the Llama 4 model. Zuckerberg had previously expressed dissatisfaction with the trajectory of the company's AI development, prompting this significant push into the frontier space.
Performance and Competitive Landscape
Independent evaluations from Artificial Analysis place Muse Wonder at a score of 52 on its Intelligence Index, while the more powerful Muse Wonder Full achieved a 56. For context, Anthropic’s Opus 5.5 currently leads the leaderboard with a score of 58. Other major competitors, including GPT-6 Astra and Gemini 4 Argon, cluster around a score of 53, placing them slightly ahead of the standard Muse Wonder model.
While Meta’s internal documentation claims the model is comparable to industry leaders, independent experts advise caution. Nick Patience of The Futurum Group noted that Meta's reported figures—such as an 85.6% score on MemoryBench—were conducted under maximum-effort conditions and should be viewed as an upper bound of performance rather than typical real-world results.
Business Strategy and Pricing
In a departure from its previous Llama strategy, Meta has opted for a closed-weight approach for the Wonder series. The standard model is available via the Meta Model API at $4 per million input tokens and $20 per million output tokens. The more capable Muse Wonder Full is restricted to subscribers of the Muse Code service, with monthly plans ranging from $50 to $200. These models were trained at Meta’s Prometheus data center in Ohio.
The commercial challenge for Meta is significant. With Anthropic’s Opus 5.5 offering higher benchmark scores at the exact same price point, Meta must prove that its infrastructure can attract developers away from established incumbents. As the company faces massive capital expenditures—projected between $130 billion and $145 billion for 2026—Muse Wonder serves as a critical test of whether Meta can successfully monetize AI beyond its traditional advertising business. Ultimately, the model's success will depend on whether it can provide enough value to justify the transition for developers in an increasingly crowded and competitive market.
Source: AI Daily News. This story was rewritten and translated for our readers.