Meta just fired its biggest shot yet in the AI race. On October 12, 2026, the company launched Muse Wonder and Muse Wonder Full — its first frontier-class AI models — from Meta Superintelligence Labs, the division Zuckerberg built over the last 14 months. “Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier,” Zuckerberg wrote on Threads. For anyone using AI to make or save money, one fact matters most: Meta is no longer the budget open-source option. It is now charging frontier prices for frontier capability.
What Meta actually announced
According to Meta’s official announcement, the Muse Wonder series represents “our most capable models to date,” built for “state-of-the-art performance on personal and professional agentic tasks.” Building on the smaller, faster Muse Spark, it targets real-world work: expert domains including law, finance, and health, plus frontier coding for people building “everything from a game to an app for a new business.”
Two features do the heavy lifting: Active Memory carries what matters about you across conversations and brings relevant details back for the task at hand, and Active Focus keeps the agent pushing toward your goal when tasks get hard — trying other approaches, asking for clarification, and checking in before consequential actions. Meta AI chief Alexandr Wang, who led the build under the internal codename Watermelon, called it “state-of-the-art at personal agents and professional agents.”
How to get Muse Wonder today
Muse Wonder is available now through the Meta Model API and inside Muse Code, Meta’s coding agent. Muse Wonder Full is restricted to Muse Code subscribers — it is not on the API at all. The consumer Muse app gets Muse Wonder “soon,” rolling out to paying subscribers first, with a 1-million-token context window.
The price tag: what Muse Wonder costs
Meta’s official post lists no prices, but VentureBeat reports via AI Weekly that Muse Wonder costs $4 per million input tokens and $20 per million output tokens on the Meta Model API — the same bracket as GPT-6 Sol and Opus 5.5. Artificial Analysis estimates $4.52 per benchmark task, nearly triple Spark 1.3’s $1.60. Muse Wonder Full runs through Muse Code subscriptions at $50/month (Power) and $200/month (Maximum). Meta spent years undercutting rivals with open-weight Llama models; that era is over — it now sells closed models at matching frontier prices.
How it stacks up — and the asterisks
On the independent Artificial Analysis Intelligence Index, Muse Wonder Full scores 56 and Muse Wonder scores 52, trailing Claude Opus 5.5 at 58. Meta’s own charts claim 57.5 on SWE-Marathon (46.3 for Opus 5.5), a reported 79.0 on DeepSWE 1.1 for coding, and a lead over Fable, Opus, and Astra on MemoryBench.
Caveat: Meta’s own newsroom post publishes no benchmark figures at all, and the launch materials reportedly omit raw per-benchmark numbers. The only scoreboard Meta does not control is starting to publish data — test before moving client work onto it.
What this means for your money
If you build with AI models, Wonder gives you a fourth serious frontier option at the standard frontier price — and live API access means you can test Meta’s claims on your own workloads. For freelancers and agencies, the practical targets are the tasks Meta names: reviewing contracts and leases, navigating insurance claims, making sense of client finances, and coding work.
Teams that built margins on cheap Meta models need to re-budget: Wonder costs roughly three times what Spark 1.3 did per task. On the consumer side, the Muse app tiers (reportedly $20 and $100 per month) put Wonder behind a paywall — and subscribers get it first.
The catches
Three caveats before you spend a dollar. First, Meta’s full safety-evaluation report is not published yet — independent verification is pending. Second, Wonder Full is locked behind Muse Code subscriptions with no pay-as-you-go path. Third, every price and most benchmark numbers come from press reports, not from Meta’s post.
Meta’s closing line: “This is a major step toward personal superintelligence for everyone, with much more to come.” The race just got a fourth serious runner — and this one owns the stadium.
FAQ
What is Muse Wonder?
Meta’s new family of frontier-class AI models, launched October 12, 2026 alongside the more capable Muse Wonder Full — built for personal and professional agentic work with persistent memory.
How much does Muse Wonder cost?
Press reports put Muse Wonder at $4 per million input tokens and $20 per million output tokens on the Meta Model API — roughly GPT-6 Sol or Claude Opus 5.5 pricing. Muse Wonder Full requires a Muse Code subscription ($50 or $200 per month).
How can I try Muse Wonder today?
Through the Meta Model API and inside Muse Code, right now. The consumer Muse app gets it soon, subscribers first.
What are Active Memory and Active Focus?
Active Memory carries what matters about you across conversations and brings relevant details back for the task at hand. Active Focus keeps the agent working toward a goal when a task gets hard — trying other approaches, asking for clarification, and checking in before consequential actions.
Is Muse Wonder better than GPT-6 or Claude Opus 5.5?
On the independent Artificial Analysis Intelligence Index, Muse Wonder Full scores 56 and Muse Wonder 52, just behind Claude Opus 5.5 at 58 — but independent verification is still pending. The real test is your own workload: the API is live, so you can measure before you commit.