Copilot Studio credits calculator: work out your Copilot credit cost, capacity packs and pay-as-you-go bill.

How to calculate Copilot Studio credit cost

Start with one interaction and count the meters it trips. A user asks your agent a question, the agent answers from a knowledge source and then updates a record. That is one generative answer at 2 credits and one agent action at 5, so 7 credits, which is 5.6 cents on a prepaid pack. Turn on tenant graph grounding for the same interaction and you add 10 credits, so it becomes 17 credits and 13.6 cents. The shape to notice is that grounding, not answering, is usually the expensive part.

Then scale it to a month and round to packs. Twenty thousand generative answers and eight thousand agent actions is 40,000 plus 40,000, so 80,000 credits. At $0.008 that is $640 of usage, but you cannot buy $640 of capacity: a pack is 25,000 credits, so you buy four packs at $200 and pay $800 for 100,000 credits. The calculator above does this rounding for you, because it is the step that separates a usage estimate from a purchase order.

Finally, subtract the traffic that does not bill. If some of those interactions come from employees who hold a Microsoft 365 Copilot license, and the agent runs under their authenticated identity, Microsoft lists those meters as no charge. Set the slider to the share of your traffic that fits that description and watch the pack count fall. For a purely internal agent in a fully licensed tenant the metered cost can approach zero, which is a genuinely different answer from the one a generic credit calculator gives you.

The zero-rating boundary that changes the whole answer

Microsoft's billing rate table has a third column that most write-ups skip, headed with whether the feature is used by a Microsoft 365 Copilot licensed user. Every metered row in it reads no charge. Read the footnote and the condition becomes precise: it covers employee-facing scenarios, described as business to employee, where the user of the agent is licensed with Microsoft 365 Copilot and the agent operates using that authenticated user's identity. Usage is subject to fair usage limits that Microsoft reserves the right to change.

Turn that around and you have the rule that decides your budget. An internal agent used by licensed staff can be close to free on the meters. The same agent, pointed at customers on your website, bills every answer, every action and every grounded lookup at full rate. Teams that pilot internally and then launch externally get a bill that looks nothing like the pilot, and the cause is not usage growth. It is a licensing boundary they crossed without noticing.

Three exclusions sit inside the inclusion and are worth knowing before you rely on it. Computer-using agents are not covered by the Microsoft 365 Copilot license and bill at the agent action rate. Agent flows are only included when the run is triggered by the agent calling the flow, so a scheduled or externally triggered flow consumes credits at the standard rate. And generative answers still incur charges unless the agent was created in Agent Builder in Microsoft 365 and the response does not use tenant graph grounding.

What happens when you run out of Copilot Credits

Microsoft allows some level of overage consumption, similar to a grace period, to avoid blocking business processes, and the Overage enforcement section of its billing rates and management article, updated on August 3, 2026, quantifies it: enforcement is triggered when a tenant reaches 125 percent of its prepaid capacity, and custom agents are then disabled. The detail that catches people is which capacity the threshold measures. Microsoft applies it to the tenant pool, so an environment holding its own allocation keeps running even while the rest of the tenant is in enforcement. This calculator therefore shows consumption against purchased capacity, which is the number the threshold is a percentage of.

The consequence that surprises people is who gets stopped. End users see agents stop responding to requests, which is the obvious half. Makers also lose the ability to use natural language to author solutions, to preview and test agents, and to generate agent evaluations. Running out of credits therefore does not just take your agent offline, it takes your development team offline too, and both happen at the same moment.

Agent flows fail differently again. Microsoft notes that when prepaid capacity is exhausted, agent flow enforcement blocks new runs rather than disabling the agent. If a customer-facing agent going quiet would cost you real money, that is worth planning around, because you will hear about it from customers before you hear about it from a dashboard unless you have an uptime monitor watching the endpoint. Your options at that point are to reallocate capacity from elsewhere in the tenant, buy more, or attach a pay-as-you-go meter to absorb the overage.

This is the strongest argument for keeping a pay-as-you-go meter linked even if you buy packs. Packs give you the better rate, the meter gives you a floor under the failure mode, and the two are not mutually exclusive.

What Copilot Studio credit pricing leaves out

The Microsoft 365 and Power Platform licensing underneath is the first omission, and for most organizations it dwarfs the credits. If the zero-rating route is the one you are counting on, you are counting on Microsoft 365 Copilot seats, and those are a separate per-user line that no credit calculation includes. Dataverse capacity and premium connector licensing sit in the same category.

Implementation is the second. Connecting an agent to real systems of record, writing and testing the actions, and putting approvals in front of anything that writes to a customer record is project work. We have not put a number on it, because any number we invented would be exactly the kind of figure this page exists to argue against.

The third is model choice inside the AI tools meter, and it moves the total more than anything else on the page. Basic prompt tools cost 0.1 credit per thousand tokens, standard cost 1.5 and premium cost 10. In dollars that is $0.80, $12 and $80 per million tokens on the prepaid rate, a hundredfold spread driven by a setting rather than by volume. If your agent uses premium reasoning tools at any scale, that line will dominate every other line, and it is the first place to look when a bill comes in higher than the forecast. The wider pattern is covered on our page about why AI agent bills exceed estimates.

The pre-purchase discount tiers Microsoft publishes, and what they actually save you

Microsoft's Copilot Credits Guide, dated August 2026, publishes a nine-tier discount ladder for the Copilot Credit Pre-Purchase Plan, which it abbreviates to P3. The committed volumes run 300,000, 1.5 million, 3 million, 15 million, 30 million, 75 million, 150 million, 225 million and 300 million credits, and the matching discounts run 5, 6, 7, 8, 10, 12, 14, 17 and 20 percent. Very few write-ups quote that ladder, and it is worth reading carefully before anyone tells you that committing is how you get a good rate.

Microsoft publishes the tiers and the percentages but not a base rate for P3, so apply each discount to the only published Copilot Credit price, the $0.01 pay-as-you-go rate, and the effective rates run from $0.0095 at tier 1 down to $0.008 at tier 9. Read the bottom of that ladder and the finding is hard to miss: you have to commit to 300 million credits, roughly $2.4 million a year, before the deepest published discount reaches $0.008 a credit. That is the same rate a $200 prepaid capacity pack gives any customer off the shelf, with no commitment at all.

The practical conclusion is that P3 is not really a volume discount, it is a budgeting instrument. What it buys is a single annual commitment covering a credit pool that spans Copilot Studio agents, Copilot Cowork, Work IQ APIs and the AI features in Dynamics 365 and Power Platform, all pooled at the tenant. If you are funding several of those in the same year, that consolidation is the value. If you just want the best price per credit, buy packs. Either way the arithmetic is set out on our guide to buying Copilot Credits, and unused P3 credits expire at the end of the annual term rather than carrying forward.

How Copilot Studio credits compare to the other vendors

On the raw rate card, Microsoft is the cheaper of the two big CRM-adjacent agent platforms per unit of work. An agent action is 5 Copilot Credits, which is $0.04 prepaid. The equivalent step in Salesforce Agentforce is 20 Flex Credits at $0.005 each, which is $0.10. That is two and a half times more for a comparable action, and you can run the same comparison against your own volumes in the Agentforce pricing calculator.

The two platforms behave in opposite ways when you overshoot, and that matters more than the rate. Salesforce states there is no overage penalty on Flex Credits and bills excess usage monthly in arrears, so you overspend quietly. Microsoft enforces capacity and stops the service, so you underdeliver loudly. Neither is better in the abstract. A customer-facing deployment usually wants the arrears model, and a cost-controlled internal rollout usually wants the hard stop.

Both currencies are also drawn down by things that do not feel like usage, which is the theme running through every metered AI product right now. If the concept itself is the confusing part rather than the arithmetic, start with what AI credits are, then read the full Microsoft breakdown on Copilot Studio pricing or the head-to-head at Agentforce vs Copilot Studio.

When a flat-rate agent is the better shape

Metered pricing is honest about one thing: it charges you most in the months the tool works hardest. That is fine when usage tracks revenue and uncomfortable when it tracks a backlog. If you run a small team and want a fixed line in the budget, the shape of the price matters as much as the rate, and no amount of forecasting makes a consumption meter predictable.

WorkAgent is $149 a month, flat, for the whole account, with no credit meter and no per-seat multiplication. We built the page you are reading, so weigh that accordingly. The honest boundary: Copilot Studio lives inside Microsoft 365 and the Power Platform, and if your data, identity and approvals already run through that stack, a general-purpose agent is not a substitute for it. WorkAgent is the better buy when the work is research, lead lists, outreach drafts, inbox triage, data entry and recurring reports, and when you want the bill to be the same in a busy month.

If you are still working out which Microsoft license you need before you get to credits at all, that question has its own answer on do you need a Copilot Studio license, and the purchasing routes are covered in how to buy Copilot Credits.