Copilot Studio Cost Per User: License Cost, Credits, and What You Actually Pay

Copilot Studio has no cost per user. Microsoft does not sell it per seat. The Copilot Studio maker licence is free of charge, and the money goes on Copilot Credits, a consumption currency pooled across your entire tenant. Buy a prepaid capacity pack of 25,000 credits a month (widely listed at $200 per pack, about $0.008 a credit), or meter it pay-as-you-go on an Azure subscription at roughly $0.01 a credit. The only genuine per-user number nearby belongs to a different product: Microsoft 365 Copilot at $30 per user per month, which zero-rates employee-facing agent usage for whoever holds it.

That answer frustrates finance teams, because a budget line needs a per-head figure. So this piece does two things: explains precisely why the per-user question has no direct answer, then shows how to derive a defensible per-person cost anyway.

Why is there no Copilot Studio cost per user?

Because Microsoft made a deliberate design choice about what to meter. Copilot Studio charges for work done, not for people present. An agent that answers 10,000 questions costs the same whether those questions came from five people or five hundred.

Two licences exist, and neither carries a price tag:

• The Copilot Studio user licence is free of charge. An admin assigns it in the Microsoft 365 admin center to someone who needs to build agents. There is a sequencing catch worth knowing: Microsoft documentation states you need a tenant prepaid Copilot Credit pack subscription before you can get this free user licence. The seat is free, the entry ticket is not.
• The trial licence is also free, and individuals can sign up for it themselves. It lets you build agents and test them in the test chat panel, but you cannot publish. Useful for evaluation, useless for production.

There is also a third route that skips licences entirely: assigning a security group to the Copilot Studio authors setting in the Power Platform admin center. Again, no per-seat charge.

So when someone asks what Copilot Studio costs per user, the accurate response is that the question is aimed at the wrong axis. Copilot Studio is priced per tenant and per unit of work. The full mechanics are laid out in our guide to Copilot Studio pricing.

What is the per-user cost of Microsoft 365 Copilot Studio agents?

If what you are really asking is how much a Copilot Studio agent costs per person who uses it, that is answerable. You divide consumption by headcount, and the arithmetic is straightforward once you know the rate card.

Microsoft charges 1 Copilot Credit per classic answer, 2 per generative answer, 5 per agent action, 10 per tenant graph grounded response, 13 per 100 agent flow actions, and 8 per page of content processing. Voice runs at 10, 35, or 75 credits per minute depending on tier.

Take Microsoft own documented sales performance agent. An average run uses four generative answers and four tenant graph grounded responses, so (4 x 2) + (4 x 10) = 48 credits per run. Across 100 chargeable users a day that is 4,800 credits daily, or roughly 144,000 a month. At the prepaid rate of about $0.008, that lands near $1,150 a month, which works out to about $11.50 per chargeable user per month.

Now change one variable. Turn tenant graph grounding off for an agent that does not need tenant-wide retrieval and the same run costs 8 credits instead of 48. Per-user cost falls to about $1.90 a month. Same agent, same users, an 83% cut from one toggle. That is why a per-user figure quoted without the agent design behind it is meaningless.

How does the Microsoft 365 Copilot licence change the per-user cost?

This is the one place a real per-seat price enters the picture, and it works in your favor.

A Microsoft 365 Copilot licence costs $30 per user per month. For anyone holding one, employee-facing agent usage running under their authenticated identity is charged at zero: classic answers, generative answers, agent actions, tenant graph grounding, and agent flow actions triggered by "When an agent calls the flow" all become no charge. Microsoft frames this as business-to-employee usage included in the user subscription licence, subject to fair usage limits it can revise.

Look again at that sales performance example. It serves 150 people, but only 100 of them generate a bill, because the other 50 hold Copilot licences. The per-user cost of the agent is therefore $0.00 for half the population and $11.50 for the rest.

The exclusions matter and are easy to trip over. Still chargeable: anything customer-facing, anything from an unlicensed user, Computer-Using Agents (explicitly carved out of the Microsoft 365 Copilot inclusion despite billing at the ordinary agent action rate), agent flows fired by any trigger other than "When an agent calls the flow", and bring-your-own-model setups such as Azure Foundry models, which bill separately.

The planning consequence is clean. For internal agents used by staff who already hold Copilot licences, the marginal cost is close to zero and you should build freely. For customer-facing agents, none of that relief applies and every turn meters.

Can I cap Copilot Studio costs at the user level?

Not per user directly, but you can get close with two controls, and you should use both before rolling anything out.

Copilot credit policies scope a prepaid capacity pack to a specific set of people. You create them in the Microsoft 365 admin center, scope each to security groups, distribution groups, or the whole tenant, and you can have up to 10 per tenant. They work standalone, with no Azure subscription and no pay-as-you-go policy needed. In practice this is how you stop one department consuming another department capacity.

Per-agent consumption limits live in the Power Platform admin center under Licensing, Copilot Studio, Manage Agents. You set a monthly credit ceiling for an individual agent so a single misconfigured one cannot drain the tenant pool.

Be deliberate about what happens at the ceiling, because the default is harsh. When a tenant hits 125% of prepaid capacity, Microsoft disables custom agents. Live conversations finish, then users get "There is a billing issue." or "This agent is currently unavailable. It has reached its usage limit." Agent flows stop earlier still, at 100% of prepaid capacity. If an agent faces customers, link its environment to pay-as-you-go billing, because environments on the Azure meter are exempt from enforcement and simply bill the overage instead of going dark.

One more trap for cost models: unused credits do not roll over. Capacity is enforced monthly and whatever you did not spend disappears. Sizing prepaid capacity for your busiest month means paying for headroom that evaporates in every quieter one, so prepaid for the baseline plus pay-as-you-go for the peaks usually costs less than prepaid for the peak.

Does the new harness change the cost per user?

It changes when you start paying, which for a pilot budget matters more than the rate.

Copilot Studio now runs on three runtimes Microsoft calls harnesses, and they bill differently. The standard harness covers rule-based agents and agent flows, and billing starts after you publish. The Copilot chat harness extends Microsoft 365 Copilot Chat and is either consumption-billed or included in the Copilot user subscription licence. The GitHub Copilot harness handles reasoning-heavy multi-step work, and it charges Copilot Credits from the moment you start building. Creating a solution in natural language, previewing, testing, and generating evaluations all consume credits, covering model tokens, tools including knowledge sources and MCP servers, and the harness runtime itself.

So on the newest harness a proof of concept has a running cost before a single user touches it, and per-user maths does not apply to that phase at all. Two teams exploring the same idea can receive very different first invoices depending only on which runtime they clicked.

Watch reasoning models too. When an agent uses one, Microsoft bills the feature rate plus the premium text and generative AI tools meter at 10 credits per 1,000 tokens. A generative answer from a reasoning model is 2 credits plus reasoning tokens, not 2 credits.

When does buying an agent beat building one?

Work the per-user number and the answer usually declares itself.

If your agent serves licensed employees and encodes rules only your company has, build it. The usage is zero-rated, the process is genuinely yours, and Copilot Studio is exactly the right tool. If your agent faces customers at volume, or does a job that is not company-specific, the calculation flips: every turn meters, you pay build credits on the newer harness before launch, and you carry maintenance forever for something that already exists finished.

Document-heavy back-office work is the clearest example. A payables agent that reads invoices, matches them to purchase orders, and routes exceptions is a genuinely common process, and content processing alone bills at 8 credits per page before any reasoning happens. Teams who only need the structured data at the end of that chain often get there faster by pointing a dedicated invoice data extraction tool at the documents and skipping the agent build entirely.

Ready-made agents in our catalog start at $29 a month with no credit pool, no monthly expiry, and no 125% cutoff, which also makes them far easier to budget: the per-user cost is the subscription divided by your team, and it does not move when someone asks the agent a harder question. Each listing sets out its capabilities, integrations, permissions, and limits so you can check the fit before deploying. Browse the agent catalog, or read our comparison of the major AI agent marketplaces if you are weighing vendor catalogs against each other.

Most organizations end up doing both: building the workflows that carry their own logic, and buying the commodity jobs so the credit pool stays reserved for work only they can do.