An AI agent costs about $20 a month to start. What you pay after that comes down to one choice

Every plan here has a ceiling. The ceilings are just made of different material, and that is what decides whether your AI agent cost stays flat or climbs.

Manus: credits, and they go fast

Manus bills credits per task, and a genuinely complex multi step task consumes far more than a simple one. That is why the tier jumps are so steep: $20 for 4,000 credits, then $40, then $200 for 40,000. Monthly plan credits do not roll over, so an unused month is money gone. Credits you buy as add-ons carry over, but only while the paid subscription is still active.

ChatGPT and Claude: usage limits, not overages

Neither one hands you a surprise bill. They throttle instead. When you hit the cap on Plus or Pro, you wait for the window to reset or you upgrade. That makes budgeting easy and capacity planning hard, because the constraint shows up as your team being blocked at 4pm rather than as a line on an invoice.

n8n: executions, with unlimited people

n8n counts one execution each time a workflow runs, regardless of how many steps are inside it. A twelve step workflow and a two step workflow cost the same. Add ten colleagues and the price does not move. Self-host the Community edition and the licence cost is zero, leaving you with a server bill and your own model API keys.

We will take a side, because "it depends on your use case" is not an answer anyone can spend money against.

For a typical five to fifty person ops team, buy two things, not one. Put $20 seats of Claude or ChatGPT in front of the people doing thinking work, and put n8n underneath for anything that repeats. That combination lands around $20 per person plus 20 to 50 euros for the automation layer, and it is the only arrangement here where adding volume does not add headcount cost.

Pick Manus when you want finished deliverables rather than a conversation. It is a strong fit when the output is a document, a deck, or a built artefact and you are willing to watch the credit meter. It is a weak fit as an everyday thinking partner, because credits make you ration the exact back and forth that makes these tools useful.

Pick ChatGPT Plus if your team already lives in it. Switching costs are real and the $20 tiers are close enough in capability that migrating for the sake of it rarely pays back. We rank Claude first for ops work, and we explain why in our Manus vs ChatGPT vs Claude comparison, but "already adopted" beats a marginal capability edge.

Do not buy the $200 tier yet. Not Manus Extended, not Pro or Max at $100 and up. Buy the $20 plan, use it hard for a month, and let the tool tell you which ceiling you actually hit. Almost nobody we work with hits it in month one, and the teams that jump straight to the top tier are usually paying for headroom they never touch.

Everything above is tool subscription cost: what the software charges you every month. It is not what it costs to have an agent designed, built and wired into your stack. Those are two different budget lines and they get confused constantly.

The subscription is almost always the smaller number. A $20 plan is trivial next to the cost of the time it takes to map a process, connect the systems, handle the edge cases and keep it running. If you are trying to budget for the build rather than the licence, we break that down separately in what you should pay for an AI agent build.

The useful sequence is this. Work out what the process costs you today. Compare that against a $20 experiment. If the experiment works and the process is worth automating properly, then price the build. Doing it in the other order is how teams end up with a signed statement of work for something a free tier would have disproved in an afternoon.