Microsoft Copilot Studio pricing starts at usage, not seats: pay-as-you-go costs US$0.01 per Copilot Credit, while the standard prepaid pack costs US$200 per month for 25,000 credits, or $0.008 per credit. The free Copilot Studio user license lets an authorized maker build and manage agents; it does not provide the tenant capacity needed to run standalone agents. Microsoft 365 Copilot can cover qualifying employee-facing use by licensed users, but it does not generally cover public website users, autonomous activity, or computer use.
For a small business, pay-as-you-go is usually the safer pilot choice. A prepaid pack becomes cheaper only when predictable billable usage exceeds about 20,000 credits per month—and unused prepaid credits do not roll over. The real budget also depends on how often an agent generates answers, grounds them in tenant data, calls actions or flows, uses reasoning models, or operates a computer.
Maker access and runtime capacity are separate decisions. Start with the channel and user entitlement, then price the billable workload.
Copilot Studio pricing at a glance
There are four different licensing ideas to keep separate: maker access, Microsoft 365 Copilot entitlements, prepaid tenant capacity, and pay-as-you-go consumption. Microsoft’s labels make these look more interchangeable than they are.
| Option | Current US list-price posture | What it covers | Best fit | Main catch |
|---|---|---|---|---|
| Copilot Studio user license | $0 | Authoring and management rights for eligible makers | Staff who build standalone agents | Does not supply runtime Copilot Credits |
| Microsoft 365 Copilot | $30/user/month, annual commitment, plus qualifying Microsoft 365 plan | Internal agent use under qualifying licensed-user identity and fair-use rules | Employee-facing agents used only by licensed staff | External users, autonomous triggers and computer use can fall outside the inclusion |
| Copilot Credit prepaid pack | $200/month for 25,000 credits | Tenant-pooled standalone Copilot Studio capacity | Stable usage above roughly 20,000 credits/month | Upfront commitment; unused monthly capacity does not roll over |
| Copilot Studio pay-as-you-go | $0.01 per credit | Actual billable credits consumed, charged through Azure | Pilots, seasonal demand and low/uncertain volume | Requires an Azure subscription and cost controls |
| Pre-purchase plan | Quote/tier dependent; Microsoft advertises savings up to 20% | Committed Copilot Credit units with PAYG overage coverage | Larger, forecastable workloads | Requires procurement and a credible forecast |
Prices above are US list prices verified August 28, 2026. Taxes, currency, agreement discounts and regional availability can change the invoice.
What does the free Copilot Studio license include?
The free user license is a permission to make and administer agents, not a free production-capacity plan. A tenant administrator normally acquires the tenant-level Copilot Studio offer and assigns the $0 user license to people who build, publish or manage agents. Publishing also requires an eligible user entitlement and available prepaid or PAYG capacity unless a qualifying Microsoft 365 Copilot scenario applies.
The free trial is useful for evaluation, and Microsoft says embedded test-chat messages are not billed. Trial rights should not be treated as a production license, however, and trial licenses do not qualify a maker to publish an agent by themselves.
This distinction matters because a small business can see “free” beside the maker license and still need an Azure-linked meter or tenant capacity before customers or unlicensed employees can use the finished agent.
How Copilot Credits are actually consumed
Copilot Credits are feature-based units, and one user request may stack several charges. The agent’s design—not merely its conversation count—determines consumption.
| Metered feature in the standard harness | Billing rate |
|---|---|
| Classic answer | 1 credit |
| Generative answer | 2 credits |
| Agent action, including a computer-use action | 5 credits |
| Tenant graph grounding | 10 credits |
| Agent flow actions | 13 credits per 100 executed actions |
| Basic text/generative AI tool | 0.1 credit per 1,000 tokens |
| Standard text/generative AI tool | 1.5 credits per 1,000 tokens |
| Premium/reasoning AI tool | 10 credits per 1,000 tokens |
| Content-processing tool | 8 credits per page |
Generative answers and knowledge grounding
A normal generative answer costs 2 credits, but optional tenant graph grounding adds 10 credits to the same response. A graph-grounded generative answer therefore costs 12 credits before any other actions or AI-tool token charges. Ordinary knowledge sources can shape the generative answer without automatically becoming tenant graph grounding; the exact configuration matters.
Agent actions, tools and flows
Actions can create several charges inside one user turn. Generative orchestration steps, triggers, topic transitions and computer-use steps are agent actions at 5 credits each. Deterministic agent-flow work is metered separately at 13 credits per 100 actions. Prompt tools add token-based consumption according to the model tier.
An agent flow does not require each maker to buy Power Automate just to create it. It is billed through Copilot Studio. A separate Power Automate cloud flow, desktop flow or process capacity remains a separate licensing and operating decision.
Reasoning models and bring-your-own models
Reasoning models stack premium token charges on top of the core feature charge. For example, a reasoning-model generative answer incurs the 2-credit answer charge plus 10 credits per 1,000 premium tokens. Microsoft-provided model rates use Copilot Credits; bring-your-own-model configurations, including Foundry models, are billed separately by their model provider or Azure resource and are excluded from the standard Copilot Credit rate table.
Four monthly Copilot Studio cost scenarios
The official estimator is most useful when its assumptions are exposed, not when it produces one reassuring total. The following models use Microsoft’s current rate table and a 30-day month. They are planning examples, not predictions of a particular agent’s production usage.
| Scenario | Transparent monthly assumption | Credits/month | PAYG cost | Cheapest standard capacity choice |
|---|---|---|---|---|
| Low-volume internal FAQ | 20 unlicensed-user sessions/day × 3 generative answers × 2 credits | 3,600 | $36 | PAYG |
| Customer-support agent | 100 sessions/day × (2 classic answers + 2 generative answers) = 6 credits/session | 18,000 | $180 | PAYG, narrowly |
| Agent calling flows/tools | 50 runs/day × [1 generative answer (2) + 2 agent actions (10) + 10 flow actions (1.3)] | 19,950 | $199.50 | PAYG, essentially break-even |
| Computer-use workflow | 20 runs/day × 12 computer-use actions × 5 credits | 36,000 | $360 | One $200 prepaid pack plus 11,000-credit PAYG overage ≈ $310, if configured |
Computer use becomes expensive quickly because each GUI action is an agent action. Real runs should be measured before committing capacity.
Low-volume internal FAQ agent
A small internal knowledge agent can remain inexpensive when its users are not already covered by Microsoft 365 Copilot and it avoids graph grounding. At 3,600 credits, PAYG is $36 per month. If all users hold Microsoft 365 Copilot licenses, use is internal, authenticated as those users, and stays within the covered conditions, the incremental Copilot Studio charge may instead be zero. Do not buy $30 seats solely to avoid a $36 meter unless the broader Copilot value justifies them.
Customer-support agent
A public support agent can approach the prepaid break-even point at modest traffic. The modeled 18,000 credits costs $180 on PAYG, still below a $200 pack. Adding one 5-credit action to only half of those 3,000 monthly sessions would add 7,500 credits and push the workload well past the pack threshold.
Agent that calls flows and tools
Flow-heavy agents need action counts, not just chat counts. In this model, ten deterministic flow actions cost 1.3 credits per run, while two orchestration actions cost 10. A standard AI tool using 2,000 tokens would add another 3 credits per run; a premium reasoning tool using the same tokens would add 20. Model choice can overwhelm the apparent savings from efficient flows.
Computer-use workflow
Computer use should be budgeted per GUI step and tested for retries. Twelve actions per run at 20 daily runs produces 36,000 credits before generative answers, reasoning tokens or failed-step retries. Computer use is not included in the Microsoft 365 Copilot user entitlement, according to Microsoft’s billing notes.
Where Microsoft 365 Copilot does and does not cover agent usage
Microsoft 365 Copilot can remove Copilot Credit charges for qualifying employee-facing usage, but it is not a tenant-wide free-runtime pass. The inclusion applies when the person using the agent has a Microsoft 365 Copilot license, authenticates with that licensed identity, and the scenario is business-to-employee.
Qualifying classic answers, generative answers, agent actions, tenant graph grounding, supported AI tools and agent-flow runs triggered by When an agent calls the flow are shown as no charge for that licensed user, subject to fair-use limits. Agent flows started by other triggers consume credits. Computer-use agents are excluded. Public websites, apps, social channels, anonymous customers, unlicensed employees and autonomous work therefore require a separate licensing analysis and usually standalone capacity.
Microsoft 365 enterprise licenses such as E3 may include the older Teams plan for creating classic agents in Dataverse for Teams. That is not the same as Microsoft 365 Copilot coverage or unrestricted standalone Copilot Studio capacity.
The costs Microsoft’s plan price does not include
Copilot Credits pay for the agent service, not the complete operating system around it. Check every layer before approving a business case.
- Azure: PAYG requires an Azure subscription; Azure AI Search, Foundry models, storage, monitoring and network services can add charges.
- Models: reasoning models add premium token consumption; bring-your-own models generate separate provider/Azure bills.
- Connectors and Power Automate: premium/custom connectors, independent cloud flows, desktop RPA, Process capacity and hosted machines may require separate entitlements.
- Dataverse: included rights and storage are finite; extra database, file or log capacity can cost more.
- Knowledge preparation: document cleanup, permissions, metadata, ingestion, search indexing and content ownership consume labor and sometimes infrastructure.
- Governance and security: environments, DLP policies, service identities, audit, Purview, managed environments and security review require administration.
- Implementation: discovery, conversation design, tool integration, testing, evaluation, exception handling and deployment are usually larger costs than the first month’s credits.
- Support and maintenance: someone must monitor failures, answer quality, consumption, connector changes and model behavior.
- Overage risk: prepaid capacity resets monthly and does not roll over; at 125% of allocated prepaid capacity Microsoft can disable custom agents or block new flow runs unless more capacity or PAYG is available.
PAYG vs prepaid: the break-even rule
At current US list prices, one $200 pack breaks even with PAYG at 20,000 credits—not at the pack’s full 25,000-credit capacity. The formula is simple:
$200 ÷ $0.01 per PAYG credit = 20,000 credits
Use PAYG when monthly usage is below 20,000 credits, volatile, seasonal or still being measured. Use a $200 prepaid pack when reliable usage is above 20,000 and below 25,000 credits, because the effective prepaid rate is $0.008. For usage above 25,000, compare additional packs, pre-purchase commitment tiers and a PAYG overage meter. Do not prepay merely because a launch forecast says “AI adoption will grow.” Forecasts are where spreadsheets go to cosplay as evidence.
Use measured monthly credits and traffic variability—not conversation counts alone—to choose capacity.
Buyer-fit verdicts for small businesses
Copilot Studio is a strong fit when a business already operates in Microsoft 365 and Power Platform, needs governed agents across internal or external channels, and can assign an owner to consumption and quality. It is less attractive when the use case is a simple deterministic automation, the company lacks Power Platform administration, or public traffic is high but hard to forecast.
- Best fit: Microsoft-centric businesses building several governed agents; internal agents for licensed Microsoft 365 Copilot users; customer agents with measurable traffic; workflows that genuinely need language interpretation.
- Proceed with PAYG first: pilots, seasonal customer service, uncertain adoption, and any agent using actions, reasoning or computer use for the first time.
- Consider prepaid: stable production agents consistently exceeding 20,000 monthly billable credits with monitoring already in place.
- Poor fit: one simple scheduled integration, unattended desktop automation that Power Automate already handles, or a tiny FAQ that can be served reliably without an LLM.
A practical capacity-planning checklist
A defensible estimate starts with a representative run, then multiplies each feature meter separately. Before buying capacity:
- Identify each channel and whether users are licensed, authenticated employees or external/unlicensed users.
- Count classic answers, generative answers, tenant graph grounding events and agent actions in a representative session.
- Count deterministic agent-flow actions separately.
- Estimate prompt-tool tokens by model tier and isolate BYO-model charges.
- Measure computer-use steps, retries and failures from realistic runs.
- Multiply by daily runs and 30 days, then add a documented buffer.
- Compare PAYG at $0.01/credit with prepaid packs at $200 per 25,000 credits.
- Add Azure, connectors, Power Automate, Dataverse, governance, implementation and support outside the credit total.
- Set budgets, alerts and an owner in Azure and the Power Platform admin center.
- Reforecast after the first production month using actual daily-aggregated consumption.
Methodology and verification
This guide is an independent analysis of current Microsoft documentation, pricing pages and the official Copilot Studio usage estimator. US list prices, feature rates, entitlement boundaries and enforcement notes were rechecked on August 28, 2026. The scenarios reproduce the estimator’s feature-based approach with their assumptions shown. No hands-on agent testing was performed, and no first-party Google Search Console or GA4 evidence was available or used.
Copilot Studio pricing FAQs
How much is Copilot Studio per month?
Standalone Copilot Studio can be billed at $0.01 per Copilot Credit through PAYG or purchased as 25,000-credit packs for $200 per month. A free maker/user license does not include production capacity.
Is Copilot Studio included with Microsoft 365 Copilot?
Microsoft 365 Copilot includes internal agent-building and qualifying employee-facing use for licensed, authenticated users. It does not broadly cover external users, autonomous triggers or computer use.
Do unused Copilot Credits roll over?
No. Microsoft says prepaid Copilot Credit capacity resets on the first day of each month and unused capacity does not carry forward.
Are agent flows included in Power Automate licensing?
No. Agent flows built in Copilot Studio are billed through Copilot Studio usage and do not require an individual Power Automate license. Separate Power Automate cloud or desktop flows follow their own licensing rules.
Does knowledge grounding cost extra?
A generative answer costs 2 credits. Optional tenant graph grounding adds 10 credits, making that combination 12 credits before other actions or tools. Other knowledge configurations should be checked against the current estimator and billing table.
What is the Copilot Studio PAYG break-even point?
At US list prices, PAYG equals the $200 prepaid pack price at 20,000 credits. Between 20,000 and 25,000 predictable monthly credits, one prepaid pack is cheaper.