Meta's Muse Wonder can drive a browser. The best version sits behind a $200 plan.

What happened

Meta announced Muse Wonder on October 12, a model built for agentic work rather than chat. Reports describe multimodal input, tool calling, and the ability to operate a browser and desktop applications. A one-million-token context window has also been reported, though that figure comes from coverage rather than Meta's own announcement page.

The part you can plan around is the price list. Consumer access is split between plans at $20 and $100 per month. The stronger version, Muse Wonder Full, is reached through Muse Code subscriptions at $50 or $200 per month.

The news here is not that Meta shipped another model. It's that the capability publishers actually want — a thing that clicks through a browser and finishes a task — is being sold as a tiered professional expense, not bundled into a chatbot.

What this is NOT

It is not a confirmed API story. Reported API pricing is $4 per million input tokens and $20 per million output tokens. That number appears in secondary coverage. We could not find it in Meta's own public materials. If you are building a cost model this week, do not anchor on it.

It is not a benchmark win you should act on. Some coverage places Muse Wonder Full fifth on a third-party index. Rankings move weekly and none of them tell you whether the model can log into your CMS without breaking something.

It is not a replacement for your editing stack. Browser control is a task-execution feature. It does not write better, and nobody has published independent reliability numbers for multi-step runs on real publisher workflows.

And it is not an argument to cancel anything yet. A $200 plan is cheaper than five tools only if it actually replaces five tools in production, with a human still checking the output.

Why the tiering is the story

For two years the pitch to creators was "generate a draft." Agentic models change the unit of work: research a topic, open the sources, pull the numbers, update the post, queue the social, check analytics. That is labor, not drafting.

When labor moves into software, the spend stops being discretionary. A $20 tool is a subscription. A $200 tool that performs a recurring Monday task is a role you decided not to hire. That's a different budget conversation, and it's the one Meta just started — the same week xAI began bundling Grok, X features and Cursor access into plans reported to run from $8 to $200 a month.

The risk nobody prices in: an agent with browser control has your permissions. Accidental publication, a wrong CMS field, a credential sitting in a session — those failures look nothing like a bad draft. They look like a correction on a live post.

What to do about it

1. Do not subscribe yet. Write down the one repeatable task you'd hand to an agent — the Monday newsletter research, the weekly analytics pull — and time how long it takes a human now.
2. Ask Meta's published plan pages directly which capabilities sit in the $20 tier versus the $100 tier, and whether browser control is in either. Coverage is inconsistent on the split.
3. Treat the $200 Muse Code tier as a replacement test, not an upgrade. List the tools it would have to retire. If the list is shorter than four, it's an expense.
4. If you trial browser control, run it in a separate browser profile with a throwaway account. Never on the profile logged into your CMS or your ad accounts.
5. Hold the reported $4 / $20 API price out of your spreadsheet until Meta publishes it.
6. Set a review date 60 days out. Agentic pricing is moving fast — Google's new tier is reportedly on promotional rates that expire November 30. Anything you lock in now gets repriced by someone.
7. Keep a log of every task the agent completed and every one it botched. In 60 days that log, not the benchmark, is what tells you whether to renew.