Manus AI Pricing: What Credits Really Cost Per Task

Manus is the closest thing on the market to what I am building. Same promise: describe an outcome, let an agent go and actually do it — browse, run code, produce the file. So I read its billing docs the way one founder reads another founder's billing docs, and the striking thing about Manus AI pricing is not the price. It is the unit. You are not buying hours, tasks or tokens. You are buying credits, and Manus will not tell you what a task costs until the task is over.

So I did the division. Here is what a Manus credit is actually worth, what that makes a real piece of work cost, and where the credit model quietly hurts.

What the money buys

Manus's help centre lists three membership tiers:

• Free — "$0/month", "300 credits provided daily", one concurrent task.
• Pro — "Starting from $20/month", "Starting from 4,000 credits per month", 20 concurrent tasks, and "Enjoy a 17% discount with annual billing".
• Pro, higher tier — from $40 a month with a 7-day free trial, from 8,000 credits a month.
• Team — from $20 per seat per month: everything in Pro plus collaboration.

A credit is not a token count. Manus's own consumption rules say a credit covers three things at once — "LLM tokens", "Virtual machines" and "Third-party APIs" — and that "Credits are only consumed during active task processing. Completed tasks and the storage or deployment of their outputs do not consume any credits." That last part is a genuinely fair design. Nobody should pay rent on a finished report.

Manus AI pricing, decoded: one credit is half a cent

Divide the entry plan's $20 by its 4,000 monthly credits and a credit is $0.005. Do it again on the higher tier — $40 over 8,000 credits — and you get $0.005 again. The unit price is flat; the tier only changes how many units you get. On annual billing at the advertised 17% off, $20 a month works out near $16.60, which puts a credit at roughly $0.00415.

Now the useful part. Manus publishes three worked task examples on its credits page:

• data analysis and visualisation, 15 minutes: 200 credits
• website design and deployment, 25 minutes: 360 credits
• app development and integration, 80 minutes: 900 credits

Multiply each by the $0.005 per credit derived above and you get $1.00, $1.80 and $4.50.

That last figure is the one worth sitting with. An 80-minute agent run costs four and a half dollars. The entry Pro plan's 4,000 credits therefore buys four of those a month with change left over — or twenty of the 15-minute analyses, or eleven website builds. Pick a genuinely busy month and $20 does not stretch as far as the headline suggests.

The free tier is tighter than it looks too. 300 daily credits will not cover two 200-credit analyses in the same day, the consumption rules cap free users' daily-refresh spend at 1,500 credits a month — about seven of those analyses — and the daily credits reset at 00:00 UTC, where "Unused credits from the previous 24-hour period are removed."

The price is fair. Not seeing it coming is the problem.

Here is the line that made me write this post. Asked whether you can check a task's cost before you begin, Manus answers: "At present, Manus does not possess the capability to autonomously judge or regulate the consumption of credits." You find out afterwards, in the usage screen.

At 200 credits that is irrelevant. It stops being irrelevant when the agent decides mid-run that this job needs forty browser pages and a build step, and you read the bill after the money is gone. It gets worse on the channel most people will actually use: Manus runs agents inside WhatsApp Business now, and its note on credits in WhatsApp Business says the WhatsApp interface does not display your real-time credit balance, so you have to log into the web workspace to check.

I am not dunking on them for it. Predicting an autonomous agent's cost before it runs is genuinely hard, and they say the product team is working on it. But "the system cannot estimate it, and the surface you chat on cannot show you the balance" is a real cost of inventing a currency, and it is the exact thing I refused to ship.

Why I price in dollars instead

Notis charges in the currency you already reason in. A plan includes an amount of usage — Pro includes $20 of usage a month — and a task draws a few cents from it. The same page publishes the actual average: last week, across every Notis user, a task cost $0.09, which is about 230 tasks inside Pro's included usage. Nothing is metered on top of the subscription; the allowance is the subscription. Pro is $13 a month billed annually.

Two consequences fall out of that choice. First, per-item tool prices are published up front rather than translated through a token of my own invention. Second, there is a hard stop built in: any single pay-per-use call quoted above $1.20 is refused, and the agent tells you the price and asks before it retries. That guarantee is not on any public page — it lives in the spend guard in the code — so take it as a description of behaviour rather than a citation.

And when the included usage runs out, the pricing page is explicit that "Nothing breaks silently" — you can route the work to your own ChatGPT or Claude subscription, keep going on on-demand usage at the same published rates and settle in arrears, or move up a plan.

Side by side

Where Manus clearly wins: it will spin up a cloud machine, write an app and deploy it, and the 900-credit example above is that capability being honest about its cost. Comparing a $4.50 app build to a $0.09 average Notis task is a unit-price comparison, not a capability one, and I would rather say that than pretend otherwise.

Which one to pay for

• Choose Manus if the deliverable is a self-contained artifact built inside a sandbox: a site, a script, an app, a deck. You are paying for compute in a VM, and credits are an honest way to sell compute.
• Choose Notis if the work is recurring business operations rather than artifacts: research that lands in Notion, follow-ups drafted in Gmail, a CRM kept current, all driven from the message thread you already live in. Dollars are the right unit when the same small task repeats four hundred times a month.
• Choose neither yet if you cannot name the task. Both tools bill for work attempted, not work you were happy with. Write down the three jobs you want off your plate first, then price them against the numbers above.

The takeaway

Credits are not a scam, they are an abstraction, and every abstraction has a bill attached. Manus AI pricing is internally consistent — half a cent a credit at both Pro tiers, and worked examples published in the open, which is more transparency than most agent vendors manage. The gap is timing: you learn the number after the spend, on a surface that will not show you the balance. Before you commit to a credit-priced agent, take the three published examples, multiply by half a cent, and ask yourself how many of those you actually run in a month. If the answer is four, the $20 plan is not a $20 plan.