Meta Claims Frontier Class AI Status with New Muse Wonder Models
Meta Platforms, parent company of Facebook, has made a significant announcement in its artificial intelligence (AI) journey, declaring that its latest model family, Muse Wonder, is among the frontier class models. This places the Muse Wonder models on par with models developed by leading AI firms like OpenAI and Anthropic.
The move marks a pivotal shift for Meta, which has historically struggled to compete with other Large Language Model (LLM) developers. Approximately 14 months ago, Meta began a major overhaul of its AI efforts, launching Meta Superintelligence Labs (MSL) and appointing Alexandr Wang as its leader. Prior to this, investors had expressed significant doubt regarding Meta’s ability to develop models that could compete with those from other LLM developers.
In April 2026, MSL released the first Muse model, Muse Spark, which Meta described as its most powerful model at the time. However, the company positioned it as “competitive” rather than as a frontier class model. With the release of Muse Wonder and Muse Wonder Full, Meta has now framed its models as truly leading edge.
According to Meta, the Muse Wonder models “deliver state of the art performance on personal and professional agentic tasks.” This claim is supported by data from model evaluation sites like Artificial Analysis, which gives the Muse Wonder Full model a score of 56, placing it among the top performers. This score ties it for second place, with Anthropic’s Claude Sonnet 5.5 also scoring 56. OpenAI’s GPT-6 Astra and Alphabet’s Gemini 4 Argon, both with scores of 53, rank just above the Muse Wonder Full. Muse Wonder, meanwhile, scores slightly lower at 52.
The cost of using the Muse Wonder models has also increased. On their max settings, Muse Wonder’s cost per Artificial Analysis Intelligence Index task is $4.52, which is 2.8 times higher than the $1.60 cost per task for Muse Spark 1.3. However, this price is still significantly lower than Anthropic’s most intelligent models and higher than those from OpenAI and Alphabet.
Notably, the more powerful Muse Wonder Full is only available through $50 per month and $200 per month subscriptions to Muse Code, not on a per token basis. This suggests that Meta views Muse Wonder Full as more suited for agentic coding tasks, while Muse Wonder is better for general purpose usage. By limiting access to Muse Wonder Full, Meta aims to create a stable, subscription based revenue stream.
Thinking about building an AI product?
Get in Touch