Meta Platforms (NASDAQ: META) has released its most powerful artificial intelligence model to date, called Muse Wonder, marking what analysts are calling the biggest comeback story in the AI industry.
The launch positions Meta as a genuine third competitor in a race that had previously been dominated almost entirely by Anthropic and OpenAI, reshaping the dynamics of the sector.
Muse Wonder is priced at $4 per million input tokens and $10 per million output tokens, matching exactly the pricing of Anthropic’s Claude Opus 5.5, making the competitive intent unmistakable.
Meta also released a more powerful Max version of the model, which competes with Fable 5.1 and GPT 6 Astra, accessible behind a $50 monthly subscription, a fraction of the $200 to $300 charged by rivals.
The model’s benchmark performance has been striking, with the Artificial Analysis Intelligence Index placing Muse Wonder in second place overall, just behind Claude Opus 5.5, after Meta’s models previously scored as low as 31 to 34 on the same index.
One area where Muse Wonder has shown a particularly dominant result is legal automation, scoring 13.1 on the Harvey Lab benchmark, more than double the second-place model, Claude Fable 5.1 Max.
Eighteen months ago, Meta’s AI ambitions appeared chaotic, having spent $25 billion to hire 29-year-old Alexander Wang, the founder of Scale AI, to lead a newly created Meta Superintelligence Lab after disbanding its previous AI team.
Two features central to the Muse Wonder strategy are Active Memory and Active Focus, with Active Memory pulling the most relevant personal context into every interaction and Active Focus enabling the model to independently find workarounds when it hits obstacles, without prompting the user.
Meta’s personal AI agent, MetaMuse, launched roughly two and a half weeks before Muse Wonder and has already accumulated 6 million downloads and approximately 3 million weekly active users, suggesting strong early product-market fit.
The strategic philosophy behind Muse Wonder stands apart from the general intelligence approach taken by Anthropic and OpenAI, with Meta explicitly targeting personalized AI designed around the individual user rather than a universally applicable model.
Mark Zuckerberg has made clear that the models Meta is releasing, including Muse Wonder and the Max version, are designed to be genuinely useful to individual users rather than broadly applicable to anyone and everyone.
Meta’s distribution advantage, with approximately 3.3 billion users across WhatsApp, Instagram, Facebook, and Threads, gives it a scale that OpenAI, still a private startup requiring significant upfront capital for compute, cannot easily replicate.
The competitive context is sharpened by the anticipated IPOs of both Anthropic, rumored to be targeting a $2.2 trillion valuation, and OpenAI, which is expected to come in at between $1.3 and $1.5 trillion, making Meta’s AI positioning a direct investment consideration.
Qualcomm (NASDAQ: QCOM) sponsored the original analysis of the Muse Wonder launch, and Alphabet (NASDAQ: GOOGL) has been identified as the only other company with comparable data and distribution assets to Meta, though it has yet to produce a competitive personal AI agent product.
Meta is also reportedly planning a more privacy-focused version of MetaMuse within the next month, a move that signals the company is aware of longstanding concerns about its data practices and is actively working to address them as part of its AI push.
