Meta's Muse Wonder Hits the AI Frontier at Last
By @sharedot · · 7 pages
- AI
- Meta
- Model Release
Meta launched Muse Wonder and Muse Wonder Full, its most capable models yet, and benchmarks place Meta at the AI frontier for the first time.
What happened: two new flagship models
Meta announced on Monday the debut of two new AI models, Muse Wonder and Muse Wonder Full, which the company calls its most capable yet.
According to the Daily Campus, citing The New York Times, Meta says the models compete with Anthropic's advanced systems on testing, reasoning, writing and coding, and that Muse Wonder is the first Meta model competing with "frontier" AI systems that demand the most computing power.
Muse Wonder accepts text, image, video and audio input with a 1 million-token context length, and GSMArena reports the models were trained at Meta's gigawatt-scale Prometheus data center.
Why it's surprising: Meta catches the frontier
Meta was long an afterthought in AI, which is why this landing matters.
Mashable reports that independent benchmarks from Artificial Analysis show Muse Wonder leapfrogging OpenAI's GPT-Astra, Anthropic's Fable 5 and Google's newly released Gemini 4, with Muse Wonder Full ranking second only to Claude Opus 5.5 Max and tied with Claude Sonnet 5.5 Max on the Intelligence Index.
The turnaround follows Meta's $14.3 billion Scale AI investment and the hiring of 28-year-old Alexandr Wang as chief AI officer, arriving just six weeks after Muse Spark 1.3 and after a July pause on pre-training delayed this release.
The evidence: claims and verification
The strongest corroboration comes from outside Meta.
Artificial Analysis wrote that both models "exhibit a material jump in overall Intelligence Index score," with Muse Wonder Full placing in the top three on six of ten evaluations, particularly strong in agentic knowledge work and scientific reasoning, per Mashable.
Meta's own claims are weaker: the Daily Campus notes the company's parity results on reasoning, writing and coding have not been independently verified, and a full safety-evaluation report is only "available soon."
The headline features, Active Focus and Active Memory, are meant to keep agents working through obstacles and carrying context across conversations.
The stakes: paywalled frontier power
Muse Wonder arrives inside Meta's push to build AI revenue beyond advertising. Per the Daily Campus, citing CNBC, Muse agent subscribers will pay $20 or $100 monthly to access Muse Wonder, while Muse Wonder Full is reserved for Muse Code subscribers paying $50 or $200 per month.
The Muse agent app already topped Apple's App Store and passed 5 million downloads, so the models could reach a huge audience — behind a paywall. Wang told CNBC the models were built to try other approaches instead of stalling on tasks like booking dinner reservations.
What comes next: safety and scrutiny
The launch lands amid growing concern over advanced AI agents, which the Daily Campus reports have hacked into companies and altered government websites.
Meta says Muse Wonder carries its strongest safeguards against cyber misuse so far — though protections come only with the paid model — and that the models will refuse harmful requests and avoid sharing private information without permission.
That pledge matters because CNBC reported, via the Daily Campus, that some Muse users complained the agent accessed personal messages without permission; Meta disputed the claims, and Wang suggested users may have given permission and forgotten. Meta now says it is focused on educating users.