Microsoft Copilot Studio pricing in 2026. What the credit meter costs you.

Copilot Studio is priced on consumption, not seats. Every agent interaction draws Copilot Credits at a published rate, so agent design and measured traffic decide the bill far more than the license does.

How does Microsoft Copilot Studio pricing work in 2026?

Copilot Studio bills on consumption. You buy capacity for the tenant, and every interaction with an agent draws it down at a published rate per feature. Since September 1, 2025, Microsoft calls the billing unit a Copilot Credit. Older contracts and invoices say messages, but the quantity per pack and the pay as you go rate did not change.

The prices sit on the Copilot Studio product page, and the rate for each feature sits in the message management documentation. Read both before you size anything, because the rate card is what turns agent design into dollars.

The credit rate card

A classic answer from a scripted topic costs the least. A generative answer, where the model reasons over your content, costs more. An agent action, where the agent plans and calls a tool, costs more again, and grounding in tenant data through Microsoft Graph costs the most per event.

The mix of these events decides the bill. One user question can trigger a generative answer, a Graph grounding call and two actions, which is 22 credits for what the user sees as a single reply.

Capacity packs, pay as you go and the prepurchase plan

There are three ways to pay, and the Copilot Studio billing documentation covers each. They differ in price per credit and in how they punish a spike.

• Prepaid packs. Tenant wide packs of 25,000 credits at $200 per pack per month, which works out to $0.008 per credit. Unused credits do not carry over to the next month.
• Pay as you go. Billed monthly through an Azure subscription at $0.01 per credit, with no commitment. You pay 25 percent more per credit than the pack rate in return for flexibility.
• Copilot Credit prepurchase plan. Commit units bought up front in the Azure portal for a one year term. Microsoft advertises savings of up to 20 percent, with deeper discounts at higher tiers.

Packs give budget certainty and the lower rate at volume. Pay as you go gives flexibility and exposes you to the full bill when an agent becomes popular inside the company.

Where Microsoft 365 Copilot changes the bill

Copilot Studio is licensed separately from Microsoft 365 Copilot, which lists at $30 per user per month on annual billing. The two overlap. When an employee holding a Microsoft 365 Copilot license uses an internal agent under their own authenticated identity, Microsoft charges no credits for answers, actions, Graph grounding or AI tools, subject to fair usage limits.

• Customer facing agents. Every event bills, because the user holds no Microsoft 365 Copilot license.
• Unlicensed employees. Staff without the license draw credits on the same internal agent that is free for their licensed colleagues.
• Agent flows. Zero rated only when the agent itself calls the flow, not when a schedule or another trigger starts it.
• Computer using agents. Not covered by the Microsoft 365 Copilot license at all.

For the wider license stack, see our Microsoft Copilot licensing guide for 2026 and our breakdown of Microsoft 365 Copilot cost per user.

The second meter in Power Platform

Agents that reach into Power Platform connectors, Dataverse or Power Automate can also draw on separate Power Platform capacity or licensing, priced on the Power Platform licensing pages. Map both meters before you set a budget. Our Power Platform licensing guide covers the second one.

What have we seen in recent Copilot Studio rollouts?

Between 2024 and 2025 we reviewed roughly 25 to 35 Copilot Studio rollouts, 30 by our engagement file count. In almost every one, consumption drove the bill. Teams budgeted for licenses and were then surprised by the meter.

The shape of the surprise was consistent. A single popular agent would consume a quarter of the monthly capacity in its first week, and finance would ask why a product sold as cheap to start was already over budget.

What does a Copilot Studio agent cost to run?

It depends on how the agent is built far more than on the price list. Two agents running on the same tenant pack can differ tenfold in monthly cost, because one answers from scripted topics and the other reaches for generative answers, Graph grounding and actions on every turn.

What drives the credit draw

• Traffic. Active users multiplied by sessions per user per month.
• Answer type. The share of generative answers against scripted topics.
• Grounding. Whether generative answers search tenant data through Microsoft Graph at 10 credits a time.
• Actions. Whether the agent triggers workflows and tools or only replies.
• Who uses it. Whether users hold Microsoft 365 Copilot licenses that remove the charge for employee use.

A worked monthly budget for four agents

The four agent profiles below come from our illustrative model. We have added hypothetical credits per session, built from the published rate card, so you can follow the arithmetic. Your own numbers must come from a metered pilot.

The procurement agent handles half the sessions of the HR agent and costs 5.75 times as much. Per session it draws 11.5 times the credits, which sits inside the range we see in live deployments.

Turning credits into a purchase

1. Convert to packs. 456,000 credits divided by 25,000 is 18.24, so you buy 19 packs, or 475,000 credits.
2. Price the packs. 19 packs at $200 is $3,800 a month, or $45,600 a year.
3. Price the alternative. The same 456,000 credits on pay as you go cost $4,560 a month, or $54,720 a year.
4. Remove zero rated use. If every HR and IT help desk user held a Microsoft 365 Copilot license and used those agents under their own signed in identity, 104,000 credits would drop out. The remaining 352,000 credits need 15 packs, or $3,000 a month.

Where do Copilot Studio budget assumptions break?

They break at the gap between a demo and production traffic. A demo runs ten sessions with a friendly tester. Production runs ten thousand sessions from staff who ask vague questions, follow up three times and trigger the expensive paths.

Why we disagree with switching it on now and sizing capacity later

The standard pitch from Microsoft and many partners is that Copilot Studio is cheap to start, so you should switch it on and size capacity later. We disagree. In most rollouts we reviewed, the cost shock came from consumption that was never modeled before launch, and a single well used generative agent could outspend a hundred seats.

The better course is to model volume per agent before you sign, cap autonomous actions during the pilot, and tie every capacity expansion to measured draw. Adoption forecasts supplied by the account team are not evidence.

Pilot draw versus production draw

Size your capacity on measured pilot draw, scaled to real headcount and real sessions per user. Never size on the capacity Microsoft suggests for a tier. In the worked example, a 45 percent overshoot turns 456,000 credits into 661,200, which needs 27 packs at $5,400 a month.

The 125 percent cutoff

Microsoft tolerates some overage on prepaid capacity, then disables custom agents once tenant consumption reaches 125 percent of the prepaid amount. Users see a message that the agent has reached its usage limit. With 19 packs, that ceiling is 593,750 credits, and the 661,200 credit month above would cross it before month end.

An agent that stops answering in week four costs more in trust than the credits would have. Environments set up with pay as you go billing absorb overage without enforcement, so decide in advance which environments get that fallback and which are allowed to stop.

The license gets you in the door. The credit meter is the bill, so model it before you sign.

How does Copilot Studio pricing change with company size?

The unit rates are the same for everyone, but what matters changes with scale. A small tenant pays mostly for minimum pack sizes and idle capacity, while a large one pays for agent design choices multiplied across tens of thousands of sessions.

A 500 person company with one or two agents

A single HR or IT agent often fits inside one or two packs. The risk is buying a pack for a peak month and wasting credits in quiet months, because unused credits expire monthly. Pay as you go is often cheaper here until draw is steady.

A 20,000 person company with a portfolio of agents

Draw can run into millions of credits a month, and the gap between $0.008 and $0.01 per credit becomes real money. The prepurchase plan, environment level capacity allocation and a firm rule on which agents may use Graph grounding and autonomous actions all start paying for themselves at this size.

How do you check your own Copilot Studio consumption?

Microsoft exposes the numbers you need, but they sit in three places. Pull all of them before a renewal conversation.

• Power Platform admin center. Under Licensing, then Copilot Studio, the Environments view shows Copilot Credit consumption per environment against what you have bought, with a details grid that breaks out agent flow actions.
• Environment capacity allocation. Manage Copilot credits shows how prepaid credits are split across environments. Unallocated credits are shared, so a test environment without its own allocation can drain production capacity.
• Agent limits. Manage Agents is where you set a monthly credit limit per agent, which is the simplest way to stop one popular agent from eating the tenant pool.
• Agent analytics in Copilot Studio. Session counts and topic usage per agent show which topics fall through to generative answers.
• Azure Cost Management. For pay as you go billing policies, set budgets and alerts on the subscription that carries the Copilot Studio meter.

What will the Microsoft account team say, and how should you answer?

Expect a small set of lines, most of them true in part. The replies below keep the conversation on measured data.

Contract terms to ask for

• Price hold on packs. A fixed pack price for the term of your agreement, so a rate change does not land midway through a rollout.
• Rate card as at signature. A reference to the credit rates per feature in force when you sign, because the rate card is what your model rests on.
• Reduction right. The ability to lower the pack count at each anniversary when measured draw falls.
• Fair usage definition. Written limits for the zero rated use by Microsoft 365 Copilot licensed users.
• Warning before cutoff. Notice to named admins well before the 125 percent point disables agents.

What are the common Copilot Studio pricing mistakes?

Most overspend we see traces back to a handful of avoidable errors.

1. Treating every agent as zero rated. Only authenticated employees with Microsoft 365 Copilot licenses qualify, so customer facing agents bill in full.
2. Leaving Graph grounding on by default. At 10 credits per event, it turns a 2 credit generative answer into a 12 credit one, often for an agent that never needed tenant data.
3. Buying packs for the peak month. Credits expire monthly, so capacity bought for a year end surge is wasted for the rest of the year.
4. Letting test environments share production capacity. Allocate capacity per environment so a developer loop cannot trigger the 125 percent cutoff.
5. Forgetting voice. A generative voice agent draws 35 credits per minute, so a five minute call is 175 credits, or $1.40 at the pack rate. That is more than 40 sessions of the scripted HR agent in the worked example.

How do you cut the Copilot Studio bill?

Four steps move the number, and each needs measured data behind it rather than a vendor forecast. They are the same steps that separated the buyers who spent less in our reviews from those who did not.

1. Model before you buy. Estimate credits per agent from a metered pilot.
2. Favor scripted paths. Route common questions to scripted topics at 1 credit per answer.
3. Cap autonomous actions. Gate the most expensive interaction type until the value per action is proven.
4. Buy packs to measured draw. Size capacity to what the pilot showed, and set a monthly credit limit on each agent.

Suggested reading

• Microsoft Copilot licensing 2026, on how the wider Copilot license stack fits together.
• Copilot Credits cost per task, with more examples of what individual agent tasks consume.
• Copilot Credits in EA negotiations, on governance and contract terms.
• Microsoft knowledge hub, the full library of our Microsoft licensing guides.

What to do next

1. This month. Run a metered pilot for each planned agent and record the real credit draw per session.
2. During the pilot. Classify every agent by answer mix: scripted, generative or autonomous, with or without Graph grounding.
3. Before sizing. Scale pilot draw to production headcount and sessions per user, then remove use covered by Microsoft 365 Copilot licenses.
4. Before launch. Route high volume common questions to scripted topics and cap autonomous actions until their value is proven.
5. At purchase. Buy packs or prepurchase units to the modeled number and allocate them per environment.
6. Before you commit. Talk to our independent Microsoft advisory team about the pack count, the contract terms and the renewal timing.

Frequently asked questions