Copilot Studio Pricing: Credits, Packs and Agent Costs
Copilot Studio credit rates, packs, pay-as-you-go and Microsoft 365 use rights, plus a worked monthly budget for an internal support agent.
Published

Microsoft Copilot Studio pricing is $0.01 per Copilot Credit on pay-as-you-go, or $200 per month for 25,000 credits, billed annually. For a staff-facing support or HR agent, check Microsoft 365 Copilot use rights first: eligible internal use by licensed employees can carry no additional agent charge. For metered use, start on pay-as-you-go and buy packs only when recurring consumption justifies the commitment.
Verified against Microsoft's live US pricing pages, October 2026; checked October 27. Prices are USD before tax. Microsoft Copilot Studio is Microsoft's platform for building and publishing agents, software that answers questions or performs tasks. The figures below price agent consumption; the Microsoft 365 Copilot pricing page covers the employee seat price.
Microsoft Copilot Studio Pricing at a Glance
The payment options buy capacity in different ways; packs and pay-as-you-go do not unlock different product features. A tenant is your organization-wide Microsoft account boundary. Pack capacity is pooled across that tenant, rather than purchased separately for every employee or agent.
The Copilot Studio plans page, Azure pay-as-you-go page, and Microsoft licensing guide supply these prices and terms. A monthly equivalent is not a monthly cancellation right: one annual pack represents $2,400 of committed subscription spending.
Microsoft Copilot Studio Pay As You Go Pricing
Pay-as-you-go charges for the credits used, with no upfront license commitment. An active Azure subscription must be linked to the relevant environment through a billing plan. The organization pays for consumption after the billing month, so a low-traffic pilot does not need a whole capacity pack.
Microsoft's licensing guide also lists $0.01 per credit, making 1,000 credits cost $10. That does not make 1,000 conversations cost $10. A conversation can contain several answers, searches and actions. Count those events before converting credits into dollars.
Copilot Studio Message Packs
The current pack contains credits, not a guaranteed number of conversations. Microsoft renamed the billing currency from messages to Copilot Credits on September 1, 2025, without changing the prepaid pack quantity or pay-as-you-go rate. The standard licensing documentation confirms that history.
At full use, $200 divided by 25,000 equals $0.008 per credit, a 20% discount against the public meter. Unused pack credits do not roll into the next month. A lightly used pack can therefore cost more per completed request despite its lower advertised unit rate.
What Each Agent Event Consumes
One answer can trigger more than one charge. Microsoft's billing rates page currently applies these event rates to the standard harness, the runtime that executes this class of Studio agent. They should not be applied wholesale to agents using the GitHub Copilot harness.
- Classic answer: 1 Copilot Credit, or $0.01 on pay-as-you-go. This is a response the maker writes in advance, such as a fixed welcome or policy notice.
- Generative answer: 2 credits, or $0.02. The model produces a response using the context and knowledge available to the agent.
- Agent action: 5 credits, or $0.05. Microsoft includes steps such as triggers, reasoning and topic transitions in this category. A business task can require several billable actions.
- Tenant graph grounding: 10 credits, or $0.10, per billed grounding event. Grounding means retrieving information to support the answer. This optional feature searches across the organization's Microsoft Graph, including external information connected to it.
- Autonomous trigger: no separate flat tariff is listed in the current rate table. Microsoft classifies triggers among agent actions, whose published rate is 5 credits per billable action. It does not publish a universal total credit price for an entire autonomous run.
For the last item, count the activity the agent performs; do not insert an assumed surcharge into every run. Microsoft's own autonomous order-processing example calculates four action calls at 5 credits each, totaling 20 credits, without adding a separate trigger fee. A scheduled HR check that performs more work needs its own activity count.
A generated answer with tenant graph grounding consumes 2 + 10 = 12 credits, costing $0.12 on pay-as-you-go, before any additional actions. A credit is a unit of consumption, not a synonym for a message, a conversation or a resolved ticket.

Copilot Studio Premium: Additional Meters to Budget
Premium reasoning adds a token charge to the underlying event charge. A token is a small unit of model input or output. Microsoft's rate card lists text and generative AI tools at 0.1 credit per 1,000 tokens for basic, 1.5 for standard, and 10 for premium. For reasoning-capable models, the premium token meter is added to the feature rate; the 2-credit generative-answer price is not the complete total.
Agent flow actions are another distinct meter: 13 credits per 100 flow actions, equivalent to $0.13 on pay-as-you-go. These are steps in a predefined automation, not the same billing category as a 5-credit agent action. Content processing is 8 credits per page. The published rates also list voice at 10 credits/minute for Classic Voice with classic orchestration, 35 for GenAI Voice, and 75 for Premium GenAI Voice, including their core answer and agent-action activity.
A text-only HR budget should exclude voice only if the design excludes voice. Similarly, a bring-your-own model, including an Azure Foundry model, is billed separately from these published Studio model rates. There is no defensible universal cost for “one HR question” without knowing the actual execution path.
What Microsoft 365 Copilot Licenses Include for Internal Agents
The end user's entitlement matters, not just the agent maker's license. The plans page includes internal agent building and use through the Copilot Chat and Standard harnesses. For standard agents, the billing conditions require employee-facing use under the authenticated identity of a user licensed with Microsoft 365 Copilot, subject to fair-use limits.
For an internal support agent, establish all of the following before removing consumption from the budget:
- The employee using it has the paid Microsoft 365 Copilot license. Owning Microsoft 365 alone is not the same entitlement.
- The agent runs in an eligible Microsoft 365 channel, such as Teams, SharePoint or Copilot Chat, with the employee's authenticated identity.
- The runtime and invoked features qualify. The standard event rates mark classic answers, generative answers, agent actions and tenant graph grounding as no-charge for qualifying licensed use.
An agent flow qualifies only when called through When an agent calls the flow during the licensed user's authenticated interaction. Microsoft's cost-management documentation excludes other flow triggers and flows started while an agent runs autonomously. It also excludes computer use and GitHub Copilot harness usage from the seat's included consumption.
For a mixed workforce, separate eligible and metered activity. Licensing the HR manager does not cover every employee who opens the agent. Nor is included usage a transferable credit allowance you can redirect toward a public support bot.
Microsoft states that fair-use limits apply but does not publish a numerical fair-use allowance in the cited billing-rate conditions. Do not translate “included” into an invented number of free conversations or promise unlimited production throughput.
The Runtime Can Change the Budget
An internal agent is not automatically an included agent. Microsoft's GitHub Copilot harness billing overview says that this runtime consumes credits for model tokens, tools and the runtime itself. Building, previewing, testing and evaluating can also consume credits, regardless of the user's Microsoft 365 Copilot license.
That overview does not supply one fixed rate for every completed task. Use its estimator and observed consumption for that runtime. The standard-harness worked example below is deliberately not a quote for it.
A Worked Month: 200 Staff and 2,000 Conversations
A 30,000-credit month costs $300 on the meter, $400 with whole packs, or $250 with one pack plus metered overflow. Those totals assume the following hypothetical staff-facing agent, with no existing spare tenant capacity and no paid Microsoft 365 Copilot entitlement covering the activity.
The company has 200 employees and expects 2,000 conversations per month. Each conversation contains one classic answer, two generative answers and one billable agent action. Half the conversations also use one tenant graph grounding event. These are planning assumptions, not measured usage or Microsoft's estimate of a typical HR agent.
The monthly count is:
- 2,000 classic answers × 1 = 2,000 credits. Include a billable opening greeting here rather than forgetting it.
- 4,000 generative answers × 2 = 8,000 credits. Follow-up answers count separately.
- 2,000 agent actions × 5 = 10,000 credits. The example assumes exactly one billable action per conversation.
- 1,000 tenant graph grounding events × 10 = 10,000 credits. These add to the answer charges.
Total: 30,000 credits, or an average of 15 credits per conversation. The example excludes premium reasoning, additional flow actions, autonomous background runs, voice, separate model charges and all other events. Add them if the design uses them.
Pay-as-you-go: 30,000 × $0.01 = $300, or $0.15 per conversation. This is the clean starting point for a pilot with uncertain demand.
Packs only: covering 30,000 credits requires two 25,000-credit packs, costing $400 per month equivalent and committing $4,800 annually. In this month, 20,000 purchased credits go unused. The effective cost is $0.20 per conversation.
One pack plus pay-as-you-go: consume the pack's 25,000 credits, then meter the remaining 5,000. The monthly equivalent is $200 + $50 = $250, or $0.125 per conversation. The pack still carries its $2,400 annual commitment. Microsoft explicitly allows pay-as-you-go after pack entitlements are exhausted.

If every employee already has the relevant paid Copilot license and every event meets the internal-use conditions, the example's additional credit charge is $0. That does not erase seat spending or implementation costs. If exactly half the modeled activity qualifies, the remaining 15,000 credits cost $150 on pay-as-you-go, below the cost of one pack.
When Does a Pack Become Cheaper?
The first pack breaks even at 20,000 metered credits per month, or 80% utilization. That is $200 divided by $0.01. Above that point it can save money, provided demand persists through the annual commitment.
The second pack has a different comparison. If you already have one pack with pay-as-you-go overflow, another $200 pack replaces up to 25,000 overflow credits. It ties when that overflow reaches 20,000 credits, meaning 45,000 total monthly credits. The two-pack-only comparison against pure pay-as-you-go ties earlier, at 40,000, but ignores the cheaper mixed option.
For any chosen number of packs, calculate the monthly equivalent as:
$200 × packs + $0.01 × credits above the packs' combined capacity.
Buy for a reliable consumption floor and meter the variable portion. A single busy enrollment month is weak evidence for a year-long capacity commitment.
Copilot Studio Licensing Guide: Annual Pools and Purchase Terms
An annual pre-purchase pool solves a different problem from monthly packs. Pack capacity resets monthly; a Copilot Credit Pre-Purchase Plan pool can be consumed across its one-year term. Its units are Copilot Credit Commit Units, or CCCUs, with 100 credits per unit.
Microsoft's published pre-purchase tiers are:
- 3,000 units: $2,850 upfront, 5% discount. This buys 300,000 credits across the term.
- 15,000 units: $14,100, 6% discount.
- 30,000 units: $27,900, 7% discount.
- 150,000 units: $138,000, 8% discount.
- 300,000 units: $270,000, 10% discount.
- 750,000 units: $660,000, 12% discount.
- 1,500,000 units: $1,290,000, 14% discount.
- 2,250,000 units: $1,867,500, 17% discount.
- 3,000,000 units: $2,400,000, 20% discount.
The smallest pool works out to $0.0095 per credit when fully consumed. Against public pay-as-you-go, its $2,850 purchase needs 285,000 credits of usage during the term to break even. This can suit seasonal demand, but the unused balance expires at the end of the year. Consumption beyond the pool moves to pay-as-you-go unless you purchase more capacity.
The separate Microsoft Agent Pre-Purchase Plan covers eligible Copilot Studio and Microsoft Foundry usage. Its published tiers are 20,000 Agent Commit Units for $19,000, 100,000 for $90,000, and 500,000 for $425,000, discounts of 5%, 10% and 15%. Each unit covers $1 of eligible usage; the term is one year, unused units expire, and overflow can be metered. A first departmental agent rarely justifies choosing a cross-service commitment before the broader usage is known.
The reviewed Studio plans and licensing pages do not advertise a general student discount or a universal refund promise. Ask the seller for the cancellation and refund conditions attached to the actual order. Do not apply consumer Copilot promotions or another Microsoft product's cancellation window to a Studio purchase.
Is Copilot Studio Free?
You can learn and prototype without buying production capacity, and some existing customers can run eligible agents without an additional Studio charge. The trial license allows standard-agent creation and testing, but not publication. The $0 maker license or author role grants access; it is not an unlimited production allowance.
For standard agents, Microsoft's billing FAQ says their own classic answers, generative answers, agent actions and tenant graph grounding in the embedded test chat are not billed. That does not establish a universal free-testing rule for every prompt tool or runtime.
There is also a narrower Teams option: classic agents created within a Teams environment using Copilot Studio for Teams do not consume credits. A standalone Studio agent merely published to Teams follows different billing rules. The channel name alone cannot establish that it is free.
Who may never need a separate Studio capacity purchase? An organization whose existing paid Microsoft 365 Copilot users only use qualifying internal agents may remain within included use rights. A department satisfied by the narrower Teams classic-agent capability may also avoid credit purchases. Neither case promises free external deployment or unattended automation.
Copilot Studio Billing: The Costs and Limits That Change the Decision
Put a usage stop on the agent if you need a spending ceiling. Allocating credits to an environment is not necessarily a cap: it can continue drawing from the tenant pool or a linked meter. Microsoft's administration guidance explicitly says Azure budgets and alerts notify you but do not stop consumption.
Three other details matter for a support desk. A proactive greeting can incur a classic-answer charge even when the visitor never replies. Preview features can still be billable. And additional storage, connected services or separately licensed automations can create expenses beyond the agent credit total. Microsoft does not publish a universal implementation fee or a fixed all-in monthly price for your workflow.
Prepaid exhaustion can also interrupt service. The standard rate documentation describes general custom-agent enforcement at 125% of prepaid capacity, while new agent flow runs are blocked when prepaid capacity is exhausted. Existing runs can finish, and the parent agent can continue non-flow interactions. Treat the overage grace as an enforcement policy, not extra purchased capacity.
For the first deployment, make these decisions before opening access:
Separate included activity from metered activity
Record the end users, their Copilot entitlement, the channel, the runtime and any unattended runs. Use the actual authenticated user for the internal-use test.
Count events for the intended workflow
Walk through a policy question, an employee-specific lookup and a request that takes action. Record answer, grounding and action counts separately. Use Microsoft's rate card for the chosen runtime, then check billed consumption after publication.
Set a monthly limit with an owner
In Power Platform admin center, open Licensing > Copilot Studio > Manage Agents. Select the agent, choose Set limit, enter the monthly credit limit, configure notifications and enable Stop usage if service should stop at that limit. Save, and assign someone to handle the alert.
Purchase against the recurring baseline
Review billed and nonbillable consumption in the environment's credit details. Start uncertain demand on the meter; add packs when the sustained baseline clears their break-even. Budget an explicit support path for employees if a limit stops the agent.
How the Cost Compares With Agentforce and Fin
Compare cost per completed request, then compare the workflow each product can support. A Microsoft credit, a Salesforce action and a Fin outcome are different units. The following normalization shows how to build a budget; it does not claim equivalent capabilities or measured resolution rates.
Salesforce Agentforce, Salesforce's agent platform, makes sense to shortlist when the work centers on Salesforce records. Its published Flex Credit price is $500 per 100,000 credits, with a standard nonvoice agent action consuming 20 credits, or $0.10. If 1,000 completed requests each require three such actions, the consumed credit value is $300. That is not the minimum invoice: packs, employee access, base subscriptions and other applicable charges still matter. The Agentforce pricing page covers that purchase in detail.
Intercom Fin AI Agent, Intercom's customer-service agent, is the more relevant shortlist candidate when the job is customer support inside a helpdesk rather than internal HR. A resolution costs $0.99, making 1,000 resolutions $990. Fin can also bill a successfully executed procedure handoff, so “billable outcome” does not always mean no human follow-up. The Fin pricing page and outcome definitions explain the boundary. The existing-helpdesk option has no Fin seat fee but has minimum commitments; your helpdesk remains a separate cost.
At the worked example's 15 credits per conversation, 1,000 Copilot Studio requests consume $150 on pay-as-you-go. Only if all those requests finish successfully is that also $150 per 1,000 completed requests. Unresolved conversations and retries still consume resources. The buying rule is therefore: prefer included Microsoft use rights when they cover the job; assess Agentforce for Salesforce-centered work and Fin for helpdesk service, then measure completion as well as consumption.
Copilot Studio Pricing FAQ
Is Copilot Studio free or paid?
Both entry paths exist. A trial lets you build and test standard agents without publishing. Production usage is either covered by specific existing use rights or paid through credits. A free maker license does not make every runtime free.
Is Copilot Studio different from Copilot?
Yes. Copilot Studio is the agent-building platform. Microsoft 365 Copilot is the employee product whose license includes certain internal Studio use rights. The seat and the agent consumption meter are separate buying decisions.
Do you need a license for Copilot Studio?
You need an appropriate access arrangement, such as the maker license associated with prepaid capacity, an author role for pay-as-you-go, a Microsoft 365 Copilot license or a trial. Production activity also needs an eligible entitlement or available billing capacity.
Is Copilot Studio included in Microsoft 365?
Paid Microsoft 365 Copilot includes qualifying internal agent use. A base Microsoft 365 subscription alone does not include every Studio capability; some subscriptions have the narrower Teams classic-agent option. Check the user's license, runtime and channel.
What are the limitations of Copilot Studio?
For budgeting, the main limits are scoped internal-use rights, monthly pack expiry, annual commitments and capacity enforcement. Trials cannot publish, and a standard-runtime estimate does not price every other runtime or separately billed service.
How much does Copilot Studio cost?
The public US meter is $0.01 per Copilot Credit. A pack is $200/month, billed annually, for 25,000 credits each month. The number of conversations that buys depends on the billable events in each conversation.
How is Copilot Studio billed?
Metered activity consumes credits at the relevant feature rates. Pay-as-you-go bills actual use monthly through Azure; packs supply monthly capacity; annual pre-purchase pools have their own term and expiry. Eligible Microsoft 365 Copilot activity is excluded from additional charges.
Which AI is better than Copilot?
For an agent purchase, choose against the workflow: Microsoft for eligible internal Microsoft 365 work, Agentforce for Salesforce-centered processes, and Fin for helpdesk customer service. A cheaper unit of billing is not proof of a better completed result.
How to use Copilot Studio for free?
Use the trial to create and test a standard agent, or use qualifying capabilities already included in your organization's licenses. Publication, unlicensed users, external channels and separately metered features can require paid capacity. Verify those conditions before inviting users.
For help turning an HR or support workflow into a scoped agent with a consumption budget, see AI agent development.
- Published
- Category
- AI
- Language










