Copilot Studio pricing: Copilot Studio licensing, cost per credit and pay as you go in 2026

Copilot Studio costs $200 per capacity pack per month for 25,000 Copilot Credits, or $0.01 per Copilot Credit on the pay as you go meter, and there is no per-user charge in either case. The pricing page describes the pack as a "tenant-wide license", and the September 2026 licensing guide adds two terms the card leaves out: the pack is "billed annually", and "Unused credits do not roll over to the next month." The Copilot Studio User License that makers need to build agents is priced at $0, and so is the Copilot Studio Author role used with the meter and the pre-purchase plans.

What the tenant actually spends is the number of credits its agents consume. Microsoft lists a $200.00 credit against a free Azure account for anyone starting out, and the pricing page says there are "no in-product features or capability differences between the Copilot Credit pack and the pay-as-you-go meter. The only difference is the way a customer pays for the service." So the purchasing decision is purely financial: a pack is cheaper per credit only if you use nearly all of it every month, and the meter is the only option with no cutoff.

One unit change is worth knowing when you read older material. The Azure retail price list shows the pay as you go meter "Pay As You Go Message" at $0.01 effective December 2024, then "Pay As You Go Copilot Credit" at $0.01 effective September 2025. The licensing guide change log confirms it: "Copilot Studio messages renamed to Copilot Studio Copilot Credits", September 2025. The price did not move; the word did, and articles that still say "per message" are quoting the old meter.

Copilot Studio is consumption priced. A pool of Copilot Credits is bought or metered at the tenant level, and each thing an agent does decrements the pool at a rate Microsoft publishes on Microsoft Learn. Every rate below is quoted from the Billing rates and management article, last updated August 3, 2026. The right-hand column is the entitlement that most cost models miss: when the person using the agent holds a Microsoft 365 Copilot license and the agent runs under their identity, employee-facing usage is "No charge", subject to fair usage limits.

Two lines dominate real bills. The first is tenant graph grounding: an agent that answers from your SharePoint, OneDrive and Teams content through the Microsoft Graph pays 10 credits for the retrieval plus 2 for the generated answer, and Microsoft uses exactly that example, "12 Copilot Credits", in the article. The second is reasoning: when an agent uses a reasoning-capable model, the premium AI tools rate of 10 credits per 1,000 tokens is billed on top of the feature rate, so a single deep-reasoning answer can cost more than a full day of classic answers.

Microsoft also publishes worked examples. A customer support agent averaging four classic and two generative answers per run at 900 customers a day bills 7,200 credits a day. A sales agent grounded in the tenant graph, answering four questions for 100 unlicensed users a day, bills 4,800 credits a day, or about 105,600 a month, which is more than four capacity packs for one hundred people.

Partly. The $30 Microsoft 365 Copilot seat includes the right to build and run internal agents in Copilot Studio using the Copilot Chat and Standard harnesses, at no additional cost, for the licensed user. The standalone Copilot Studio license is required to publish agents to external channels, to let unlicensed people use them, and to build with the GitHub Copilot harness. That is the pricing page's own FAQ answer, condensed, and the licensing guide's use-rights table says the same thing in more detail.

The distinction that decides your bill is who is using the agent, not who built it. The billing rates footnote reads: "Employee-facing usage scenarios (Business to Employee) of Copilot Studio agents and Copilot Studio tools and features that these agents invoke are included in the Microsoft 365 Copilot USL when the user of the agent is licensed with Microsoft 365 Copilot, and the agent operates using the authenticated Microsoft 365 Copilot USL user's identity." So a company where every employee already pays $30 a month runs internal agents on that seat, within fair use. A company that licenses Copilot for 200 people and points an agent at 2,000 pays credits for the other 1,800.

Two exceptions in the current guide are new. Work IQ API usage by Copilot Studio agents is "not included as entitlements in the Microsoft Copilot license" (July 2026), and computer-using agents are billed at the agent action rate even for licensed users. There is also a naming split you will hit while reading: the guide's change log states "Microsoft 365 Copilot rebranded as Microsoft Copilot" in August 2026, while the pricing page on September 21, 2026 still sells "Microsoft 365 Copilot" at $30. Both names refer to the same $30 seat.

Pay as you go costs less whenever a tenant uses fewer than 20,000 of a pack's 25,000 credits in a month, because $200 divided by 20,000 is the meter's $0.01. Above 80 percent utilization the pack wins, up to a floor of $0.008 per credit at full use. The utilization test has to hold every month, because the guide is explicit that pack credits do not roll over and that the pack is billed annually. A team whose usage swings between 10,000 and 30,000 credits pays for the peak in packs and loses the trough.

The pre-purchase plans change the arithmetic for uneven usage. The Copilot Credit Pre-Purchase Plan discounts the meter by tier: 5 percent at 300,000 credits, 6 percent at 1.5 million, 7 percent at 3 million, 8 percent at 15 million, 10 percent at 30 million, 12 percent at 75 million, 14 percent at 150 million, 17 percent at 225 million and 20 percent at 300 million. The credits last the full year and "If usage exceeds the available number of credits, customers can purchase additional Copilot Credits or be billed on a pay-as-you-go basis." At tier 1 the effective rate is $0.0095, worse than a fully used pack but better than any pack used below 84 percent, with none of the month-end loss. The Microsoft Agent Pre-Purchase Plan, added in February 2026, works the same way in Agent Commit Units worth $1 or 100 credits each, at 5, 10 and 15 percent off for 20,000, 100,000 and 500,000 ACUs, and can be spent on Microsoft Foundry as well as Copilot Studio.

The safest configuration for a production agent is a pack or prepaid pool with a pay as you go meter linked behind it. Microsoft Learn describes the pairing directly: prepaid credits are used first, and "any usage beyond the prepaid capacity amount is automatically billed at the per-credit pay-as-you-go rate" with "no service interruption." Without that link, the enforcement rules in the next section apply. The break-even at 10,000 to 300,000 credits a month is worked through in Copilot Studio pay as you go vs capacity packs.

On prepaid capacity, agents are switched off. Microsoft Learn states: "Enforcement is triggered when a tenant reaches 125% of their prepaid capacity" and, on the action taken, "Custom agents are disabled. Disabling an agent doesn't interrupt an ongoing conversation. All subsequent attempts to invoke the agent are rejected until capacity is increased or reset." End users then see one of two messages, quoted in the article: "There is a billing issue." or "This agent is currently unavailable. It has reached its usage limit." An email goes to the tenant administrator and a notice is posted in the Power Platform admin center.

Agent flows are stricter. Their enforcement is separate and starts at 100 percent, not 125: "When a tenant's prepaid Copilot Studio capacity is fully consumed, new agent flow runs are blocked." Runs already in progress complete, the parent agent keeps answering, and the block resets when prepaid credits renew monthly. Test runs and flows invoked by Microsoft 365 Copilot licensed users through the "When an agent calls the flow" trigger are exempt.

Three fixes are documented: reallocate capacity between environments in the Power Platform admin center, buy more credits, or link an Azure subscription so overage bills to the pay as you go meter, at which point "enforcement doesn't apply because overage is billed to your Azure subscription." Admins can also cap individual agents under Licensing, Copilot Studio, Manage Agents, so one runaway agent cannot drain the tenant pool. Microsoft's allocation example is worth reading before you design environments: an environment with its own allocated credits is protected from a tenant-wide overage, and an environment on pay as you go is never enforced at all.

For the most common Copilot Studio use, an internal agent that answers employee questions from company content through tenant graph grounding, the meter works out to about $11 to $13 per employee per month at five questions a day, and about $22 to $26 at ten, for people who do not hold a Microsoft 365 Copilot seat. The model below uses Microsoft's own 12 credits per grounded answer (10 grounding, 2 generative), 22 working days, and no agent actions, flows or reasoning, so it is a floor rather than an estimate of a real deployment.

Read it honestly in both directions. At light use over Microsoft 365 content only, Copilot Studio undercuts a $20 per seat search tool, and a company that already pays $30 a seat for Microsoft 365 Copilot runs the same agent on that seat for no extra credits. The number crosses $20 a head at roughly eight grounded questions per person per day, and it climbs faster than the table shows once an agent uses actions, flows or a reasoning model, because those meter on top of the 12. What the table cannot price at all is content outside the Microsoft Graph: an agent grounded on SharePoint does not see Slack, Notion, Jira or Google Drive unless that data is first brought into the Graph, and every one of those is a build and a governance decision rather than a line item.

That is the boundary to decide on before choosing a purchasing option. If the questions your people ask are answered inside Microsoft 365 and usage is light, buy packs with a meter behind them and stop reading. If the answers are spread across Microsoft 365, Google, Slack, Notion, Confluence and Jira, and people will ask all day, a flat per-seat index with cited answers costs less and needs nothing built. That is what InSearch is: $20 per seat per month on the Team plan, every connected source indexed once, permissions checked per person at query time, and no meter that switches anything off.

Yes. Anyone with a work or school account can start a Copilot Studio trial, extend it once by 30 days when it expires, and keep a built agent running for up to 90 days after expiry. The trial cannot publish: Microsoft Learn states that "The trial license gives you access to Copilot Studio to create agents. You can test your agents using the test chat panel. However, you can't publish the agent." Personal email addresses are rejected, and an IT administrator can disable self-service sign-up for the whole tenant.

Separately, the pricing page offers a $200.00 Azure credit with a new free Azure account, which can fund the pay as you go meter for 20,000 credits of real, published usage. For a licensed pilot the licensing guide points to the Copilot Studio for Teams plan included in select Microsoft 365 subscriptions, which supports classic orchestration and publishing to Teams only, with no generative orchestration and no premium connectors. Microsoft also maintains a public Copilot Studio agent usage estimator for forecasting credits by agent type, traffic, orchestration, knowledge and tools before you buy anything.

Copilot Studio is priced for building agents, and it prices that well: the rate card is public, the meter has no minimum, and licensed users are not double charged. It is the wrong purchase when the job is simply finding answers across every system a company uses, because the meter charges per question, the grounding stops at the Microsoft Graph, and someone has to build, allocate and govern the agent in Power Platform.

Buyers usually arrive at this page from one of two places. The first is a licensing review, where the question is whether the $30 seat, the credit meter or both apply; the licensing guide and the tables above answer that. The second is a make-or-buy decision for an internal knowledge assistant, where the honest comparison is a credit forecast against a per-seat price. Run that forecast on Microsoft's estimator with your real question volume and your real sources. If the sources are all in Microsoft 365 and the volume is low, Copilot Studio is the cheaper answer. If they are not, connect them to InSearch instead and ask a question against your own content before you commit to either.