Microsoft 365 Copilot cost and Azure OpenAI licensing. Two pricing models, one AI budget.

Microsoft sells AI two ways at once: per user for Copilot and per transaction for Azure OpenAI and Copilot Studio. Most budget errors come from mixing up the two models or forecasting the meters on the wrong rate.

How much does Microsoft 365 Copilot cost, and how is the rest of Microsoft AI priced?

Microsoft 365 Copilot lists at $30 per user per month on an annual commitment paid yearly, or $31.50 per user per month if you pay monthly on the same annual term. Every seat also needs a qualifying Microsoft 365 or Office 365 base license, which is a separate line in the budget.

The rest of Microsoft AI follows one of two pricing models. The Copilot products are sold per user per month. Platform services such as Azure OpenAI are consumption based, metered per token or per unit of reserved capacity. The two rarely appear on the same quote, which is why budgets so often miss one of them.

Our Microsoft 365 Copilot pricing guide tracks list price changes and promotions as they happen.

Which base licenses qualify for Copilot?

The Copilot add on cannot stand alone. Each seat needs a qualifying base such as Microsoft 365 E3 or E5, Business Standard or Premium, or Office 365 E3 or E5. Microsoft sets the eligible bases on the Microsoft 365 Copilot page and publishes the full list in the Copilot licensing documentation.

• Enterprise bases. Microsoft 365 E3, E5, F1 and F3, Office 365 E1, E3, E5 and F3, and Microsoft 365 Apps for enterprise. Microsoft 365 E7 already includes Copilot, so it needs no add on.
• Business bases. Business Basic, Standard and Premium, and Microsoft 365 Apps for business. These tenants can also buy Copilot Business instead of the enterprise SKU.
• Standalone workloads. Exchange, SharePoint and OneDrive plans, the Teams Enterprise and Teams Essentials plans, and Project, Planner and Visio plans also qualify.

Check frontline users with care. An F1 or F3 user can hold Copilot, but many of those roles do not work in Word, Excel or Outlook enough to justify $360 a year. Our guide to Copilot license prerequisites covers the government and education variants.

Does Microsoft 365 E7 make Copilot cheaper?

E7 lowers the cost of Copilot only if you would buy everything else in the bundle anyway. Microsoft 365 E7 lists at $99 per user per month paid yearly and includes Microsoft 365 E5, Microsoft 365 Copilot, Agent 365 and the Entra Suite. Microsoft 365 E5 lists at $60, so E5 plus a $30 Copilot seat comes to $90.

The $9 gap buys Agent 365 and the Entra Suite for every user, or $108,000 a year across 1,000 users. E7 also puts Copilot on every E7 user, so you lose the option to license it only where it earns its cost. Our Microsoft 365 E7 guide covers when the bundle pays back.

How is GitHub Copilot licensed?

GitHub Copilot is a separate per user product with Business and Enterprise tiers. The tiers differ on policy controls, knowledge features and administrative reach. Current pricing sits on the GitHub Copilot plans page, and organization quotes come through GitHub sales.

• Business. Organization wide controls at $19 per user per month.
• Enterprise. Deeper administrative reach and knowledge features at $39 per user per month. It requires GitHub Enterprise Cloud.
• AI credits. Since June 1, 2026, premium requests have been replaced by GitHub AI Credits, billed on token consumption at each model's rate. Business includes $19 of credits per user per month and Enterprise $39, pooled across the organization, and unused credits expire at month end. Usage beyond the pool is billed at $0.01 per credit, so set organization and cost center budgets in the billing settings.
• Separate budget line. GitHub Copilot does not draw on Microsoft 365 Copilot seats, and a developer who uses both needs both.

More detail on the tier choice sits in our GitHub Copilot Enterprise licensing guide.

How does Azure OpenAI Service billing work?

Azure OpenAI charges per token, split into input and output rates, with the output rate usually several times higher. Older price sheets quoted rates per 1,000 tokens. The Azure OpenAI pricing page now quotes them per 1 million tokens, so check the unit before you copy a rate into a model.

Three details change the bill more than the headline rate. Cached input tokens are priced below fresh input on the models that support caching. The Batch API runs jobs within 24 hours at a 50 percent discount on Global Standard pricing. Heavy, steady traffic can move to reserved capacity instead of the token meter.

Worked example: what happens when output tokens are left out of the forecast

Say you run an internal assistant that handles 50,000 requests a day, with 1,500 input tokens and 300 output tokens per request. The rates below are illustrative, $2 per million input tokens and $8 per million output tokens. Use the published rate for the model and deployment type you actually run.

The forecast misses by $2,700 a month, or 50 percent, although output is only one sixth of the token volume. Long answers, summaries and code push the output share and the gap higher. Any part of the load that can wait up to 24 hours, such as overnight document processing, costs half as much through the Batch API.

When should you buy Provisioned Throughput Units?

Provisioned Throughput Units reserve dedicated capacity for predictable latency at a fixed price. Microsoft explains the model in the provisioned throughput documentation. PTUs bill hourly by default, and an Azure reservation for a one month or one year term lowers the hourly rate.

• Pay as you go or provisioned. The meter suits pilots and variable workloads. Provisioned throughput suits steady, high volume production traffic once it beats the pay as you go rate, priced with output tokens at the higher rate.
• Reservations do not guarantee capacity. Create the deployment first to confirm capacity exists in your region or data zone, then buy the reservation.
• Minimums vary by model. Each model has a minimum PTU count per deployment, so a small workload can be too small to reserve.

Spillover routes overflow requests from a fully used provisioned deployment to a standard deployment, so you can size the reservation to the baseline rather than the peak. For how Azure OpenAI spend interacts with an Azure commitment, see our Azure MACC negotiation guide and the Azure OpenAI enterprise pricing guide.

What does Copilot Studio cost?

Copilot Studio bills on Copilot Credits, the unit Microsoft called messages until September 1, 2025. A prepaid pack gives your tenant 25,000 credits a month for $200, and a pay as you go meter billed through Azure handles overflow at $0.01 per credit. Unused pack credits do not carry over to the next month.

The cost driver is the weight of each event. A classic answer costs 1 credit, a generative answer 2, an agent action 5 and tenant graph grounding 10. Generative answers and multi turn agents therefore consume far more than a simple scripted bot, so volume estimates drive the budget.

How do Copilot Credit packs compare with pay as you go?

A pack suits predictable volume and a steady run rate. Pay as you go suits early pilots where volume is unknown. Most enterprises start on the meter, then move the stable baseline onto packs once usage settles.

A third option is the Copilot Credit Pre-Purchase Plan, a one year commitment bought through Azure. Microsoft's own example shows a 6 percent saving against pay as you go at the second tier. The purchase cannot be canceled or exchanged and renews automatically by default, so commit only volume you have already seen on the meter.

Worked example: sizing credits for an HR agent

Say an HR agent handles 20,000 conversations a month from employees without a Microsoft 365 Copilot license. Each conversation averages three generative answers and one agent action, so 3 × 2 + 5 = 11 credits. That is 220,000 credits a month, or $2,200 on the meter.

Nine packs cover 225,000 credits for $1,800 a month. If the real volume turns out 30 percent lower, you pay for credits that expire.

What have we seen in recent Microsoft AI negotiations?

Across roughly 25 to 35 Microsoft AI engagements we advised between 2024 and 2025, the first budget was usually built on list price seat counts with no usage forecast behind it. The same three patterns came up repeatedly.

• Seats ahead of use. Microsoft 365 Copilot was bought for 2 to 5 times the number of users who reached steady weekly use within the first quarter.
• Token forecasts built on input only. Azure OpenAI spend ran 20 to 60 percent above forecast where teams priced on input tokens and ignored the higher output rate.
• Agent volume underestimated. Copilot Studio volume came in 30 to 50 percent above estimate once generative actions and multi turn agents went live.

In one engagement we cut a planned Copilot order from 1,200 seats to 400, instrumented weekly active use, and expanded only against proven adoption. The first year cost fell by more than half. At list price, 1,200 seats commit $432,000 a year, while 400 seats commit $144,000.

Why we advise against licensing Copilot for everyone on day one

The standard reseller pitch is to license Microsoft 365 Copilot broadly on day one so the whole organization can adopt it together. We think that is the wrong order. The 2 to 5 times overbuy above came from exactly that approach, and it stranded money on an annual commitment that cannot be reduced mid term.

License a measured pilot of the roles with the heaviest document and meeting load, and give everyone else the Copilot Chat access their existing subscription already includes. Measure weekly active use for 60 to 90 days, then expand in tranches as each group shows steady use. Our Copilot pilot program describes how we set one up.

How do you check what your organization actually uses?

Each product has a native report. Pull all four before any renewal or expansion talk, so you hold a usage number per product that you trust more than the account team's.

• Microsoft 365 Copilot. The Microsoft 365 Copilot usage report in the Microsoft 365 admin center shows enabled and active users by app. The Copilot Dashboard in Viva Insights adds trends by group. See our note on measuring Copilot active users.
• GitHub Copilot. The Copilot seat assignment list in the organization settings, and the REST API, show the last activity date per seat. Seats idle for 30 days or more are candidates for removal.
• Azure OpenAI. Azure Cost Management split by meter shows input, cached input and output spend separately. For PTU deployments, the provisioned throughput metrics in Azure Monitor show how busy the reserved capacity is.
• Copilot Studio. The Copilot Studio capacity view in the Power Platform admin center shows credit consumption by environment and agent against your packs.

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

Expect the conversation to push toward broad seat counts and multiyear commitments. These are the lines we hear most often, with the reply that holds up.

A Copilot seat is a monthly cost that has to earn its place, and the meters behind Azure OpenAI and Copilot Studio have to be forecast before anyone signs.

Which contract terms should you ask for?

Ask for these terms before signing any Copilot, Azure OpenAI or Copilot Studio commitment. Most of them protect your right to change volume later, which is worth more than a few extra points of discount while adoption is still unproven.

• Tranche pricing. The Copilot price for later expansion tranches fixed now, so a successful pilot does not reset the discount.
• Reduction rights at anniversary. The right to reduce Copilot seats at each anniversary written into the order, whichever channel you buy through, because an annual subscription cannot be reduced mid term.
• Price hold. A hold on the Copilot and base license prices for the full term, since Microsoft has already raised E3 and E5 list prices.
• Consumption eligibility. Written confirmation that Azure OpenAI and Copilot Credit Pre-Purchase spend count toward your Azure commitment.
• Pilot credits. Funded pilot or deployment credits for the first tranche rather than a deeper discount on seats you may not use.

What to do next

1. Map the use cases. List every AI use case and tag each as per user or consumption before pricing anything.
2. Check the base. Confirm the qualifying base license under every intended Microsoft 365 Copilot seat.
3. Pilot first. Run a measured Copilot pilot and instrument weekly active use before any broad order.
4. Forecast output tokens. Forecast Azure OpenAI tokens with the higher output rate, not the input rate alone.
5. Estimate agent volume. Estimate Copilot Studio credit volume with generative actions switched on.
6. Reserve late. Move steady Azure OpenAI traffic to provisioned throughput only once volume is proven.
7. Review quarterly. Re forecast every quarter and release seats and capacity that adoption did not justify.

Frequently asked questions