What Muse costs — and what we'd actually buy
Meta's paid-tier numbers, checked 3 November 2026 · Our read below them
Meta's numbers, stated once
Meta Muse has three plans: Free ($0, with a weekly usage limit), Power ($20 a month for 500M tokens a week) and Maximum ($100 a month for 3B tokens a week). The paid-tier figures were checked against Meta's own subscription help page on 3 November 2026; the free-tier figure is Zuckerberg's launch statement, not a Meta-published number.
| Plan | Price | Tokens / week | Where to get it |
|---|---|---|---|
| Free | $0 | 100M* | muse.ai or the Muse app — iOS, Android, web, Mac, WhatsApp (US). Monthly, cancel anytime. |
| Power | $20/mo | 500M | |
| Maximum | $100/mo | 3B |
* Meta's subscription help page publishes no free-tier number — the 100M tokens/week figure comes from Zuckerberg's launch statement.
We don't control these numbers and Meta can change them — if anything here disagrees with your app, Meta's numbers win. Everything below the table is our own read: what the tiers mean in practice, which one to buy, and what Meta's page doesn't say.
What is a Muse token, honestly?
Meta has not published a definition. There is no official conversion from a Muse token to messages, tasks, minutes or words of compute — so nobody, us included, can tell you that 100M tokens equals a specific number of errands a week. We are not going to invent conversion math to make a table look tidy.
Practical approach: use Free for a week and watch how you get on. Your own usage is the only converter that exists. If you're choosing between assistants rather than tiers, see Muse vs other agents.
Why work stops mid-task, and how to make tokens last
A task that stops halfway is not one thing. It can be the weekly allowance — but it can just as easily be the surface you're on, a connector that dropped, a chat that's grown too long, an approval you never answered, or a site that refused the agent. Without a bit of context you can't tell which one hit you. Before you assume it was tokens, check four things: which surface was it? (approvals and the full audit trail live in the app and web; WhatsApp stalls more often and shows less), which connectors did it need? (one expired OAuth token or a disconnected account ends a task mid-step just as cleanly), how old is that chat? (Muse carries memory and thread history into every request, so a week-old chat makes every step more expensive and more likely to drift), and what was the last step? — a connector read, a browser page, or an approval card waiting for you.
If those four come back clean and the meter is spent, then it is the allowance: the weekly budget is spent while the work runs — reading the thread, opening the page, writing the draft, checking the result — and when it runs out the task stops where it is, sometimes without a warning first. Meta publishes no per-task conversion, so treat any "one errand equals X tokens" figure you see quoted as somebody's guess. What you can actually control is how much Muse has to read to finish the job — and that is where most of the difference between a cheap week and an expensive one comes from.
Why tasks actually stall
- The surface. WhatsApp is quick for small asks but has no full audit trail and stalls more often; approvals live in the app and web. A task that "mysteriously stopped" on WhatsApp is often waiting on you.
- A dropped connector. Expired OAuth, a disconnected account, a scope you revoked — the step that needed it fails and the task ends there. Check Muse → Connectors before anything else.
- An unanswered approval. Anything that sends, spends or shares waits for your yes. No answer, no next step — it looks like a stall, not a refusal.
- A blocked step. Login walls, verification codes Muse filters out by design, and sites like Amazon that refuse agent traffic all stop the job dead.
- Memory and thread weight. Muse carries what it remembers — chats, goals, connector context — into every request. A chat that's run all week gets slower and pricier with each step. Start a fresh chat when the subject changes.
- The allowance. If the weekly meter is spent, nothing above matters until it refreshes — or you upgrade. When it runs out, work stops where it is.
Make the same budget go further
- One task per chat. Clean context, nothing re-read from last week.
- State the finish line. "Summarise the last 20 unread into urgent / can wait / newsletters" beats "deal with my inbox".
- Name the source. Point it at the connector ("my Google Calendar") rather than letting it browse to the same answer.
- Ask for the summary, not the dump. "Don't quote the thread — just tell me what changed."
- Interrupt loops. If you see the same step twice, stop it and rephrase; that is where a week disappears.
- Put the heavy job after the refresh. Watch for a couple of weeks and you will learn which day your allowance resets.
If it stops halfway through
- Ask the four questions first. Which surface? Which connectors? How old is this chat? What was the last step? That usually names the cause — a stuck approval, a dropped connector, a WhatsApp stall, a weighty thread — before tokens even come up.
- Then check the meter. If the allowance is spent, waiting for the refresh or upgrading is the only fix — rephrasing will not buy tokens.
- Resume with state, not "continue". Say what is already done and what is left, and on which surface. A bare "continue" — especially in a long thread — can send it back over reading it has already paid for.
- If it is time-critical, finish the last step yourself. A booking that has to happen today should not wait on a weekly budget or a flaky connector.
If you are hitting the cap most weeks, the honest fix is a bigger allowance, not shrinking every task — the buying advice below covers which tier actually fits.
Which tier should you actually buy?
The short answer: start on Free and stay there until the weekly cap actually interrupts you. Power ($20 a month) is the right tier for most people who outgrow Free — daily use, at least one real connector, capped at least twice. Maximum ($100 a month) is only for people who capped Power two months running or run long autonomous jobs.
Stay on Free
You're still deciding whether this is useful. A few tasks a week, a few connectors, and you've hit the cap fewer than twice. Give it a fortnight before spending anything.
Go Power
You use it most days, you have at least one real connector, and the cap interrupted you this month. This is the right answer for most people who read this far.
Consider Maximum
You capped out on Power two months running, or you run long autonomous jobs against several sources. If neither is true, wait.
Nobody ever regretted starting on Free and buying the upgrade one week late. Plenty of people regret the reverse. When you hit the weekly limit, Meta's guidance is simple: upgrade or wait for the allowance to refresh — note the reset day appears to vary by account rather than being fixed, so confirm it in your own app.
The costs that aren't on Meta's pricing page
The subscription is only the first cost. The subscriptions of the apps you connect, the time you spend on permissions, where you bought the plan if you want a refund, and invite bonuses that expire — none of these appear on Meta's pricing page, and all of them are real money or real minutes.
- The attached services. Spotify Premium, Google One, Notion — Muse connects to them; Muse doesn't pay for them.
- Your time on permissions. Five minutes on privacy settings and the audit habit from the permissions guide are part of the real cost of running an agent safely.
- Refund windows. Subscriptions bought on mobile storefronts are often refunded through Apple or Google rather than Meta, which changes how you get your money back.
- Invite bonuses that expire. Reported as usage credit with an expiry rather than subscription credit, and they can't be cashed out — see the invite code page.
Covered on their own pages
Pricing touches three topics big enough for their own pages: invite-code bonuses, how paying inside Muse works, and what a team rollout costs. Each is covered in full below.
- Invite codes — what the reported 1B-token bonus is and which offers to ignore: invite code guide →
- Paying inside Muse — Link by Stripe, single-use cards, purchase protections: is Muse safe? →
- Pricing a team rollout — seats, licences, the policy hour, pilot caps: Muse for small business →
Pricing: questions
- How much does Muse cost?
- Free with a weekly usage limit — Zuckerberg said up to 100M Muse tokens a week at launch, but Meta's own help page publishes no free-tier number. Power is $20 a month for 500M a week, Maximum is $100 a month for 3B a week. Monthly, cancel anytime. Paid tiers checked against Meta's subscription help page on 3 November 2026 — if your app shows different numbers, Meta wins.
- Can I cancel anytime?
- Yes — the plans are monthly and Meta describes them as cancel whenever you like. You drop back to the free tier's weekly allowance rather than losing the account.
- Is there a student or business discount?
- Not published. Meta's subscription help page lists Free, Power and Maximum with no student, education or business pricing tier, and we are not going to invent one.
- Why did Muse stop in the middle of a task?
- It is not always the token allowance. Most mid-task stops are one of: the surface (WhatsApp shows less of the audit trail and stalls more than the app or web), a dropped or expired connector, an approval card waiting for your answer, a blocked step (login walls, codes Muse filters out, sites that refuse agents), a chat that has run all week and carries too much history, or — if the weekly meter is spent — the allowance itself. Check those in that order before assuming it was tokens.