Microsoft Copilot Credits Magic Map

A practical licensing and consumption guide for navigating included experiences, metered agentic work, Cowork, Work IQ APIs, and separate capacity models.

Microsoft’s Copilot portfolio now combines subscription entitlements with usage-based services. That flexibility is useful, but it also makes a simple question surprisingly difficult: Will this interaction be included, consume Copilot Credits, or land on a different bill? This guide turns that question into a repeatable decision process for September 2026.

Licensing noteThis article is practical guidance, not a substitute for Microsoft Product Terms or a customer’s CSP or Enterprise Agreement. Pricing is in USD, is subject to change, and contractual terms prevail. Validate region, cloud, preview status, and purchasing terms before placing an order.

Executive summary

Copilot Credits are a tenant-pooled consumption currency, not a per-user license. Pay-as-you-go is listed at $0.01 per credit, with capacity packs and annual pre-purchase options available for other consumption profiles.
A Microsoft 365 Copilot user subscription does not include Cowork consumption. The subscription is a prerequisite, while Cowork tasks consume credits according to the work performed.
Normal licensed Microsoft 365 Copilot productivity use is included. Core chat, supported Microsoft 365 app experiences, native Work IQ grounding, and qualifying Microsoft-built experiences do not create incremental Copilot Credit charges.
Agent identity and execution context matter. Authenticated business-to-employee use can be included for a licensed user, while anonymous access, external channels, autonomous execution, certain harnesses, and advanced capabilities can be metered.
GitHub Copilot is a different product on a different meter. It is licensed per developer seat, includes GitHub Copilot Chat with a paid seat, and consumes GitHub AI Credits rather than Copilot Credits.
Security Copilot is separate. Security Copilot uses Security Compute Unit capacity rather than Copilot Credits.

🟩 Included — no incremental Copilot Credit charge🟨 Copilot Credits — metered, funded by PAYG, packs, or P3🟥 Separate meter — not funded by Copilot Credits🟦 Varies — depends on the product entitlement invoked

Note that 🟨 Copilot Credits describes the meter, not the purchasing route. The same metered usage can be funded by pay-as-you-go, a capacity pack, or a pre-purchase plan.

The consultant’s 30-second magic map

Step 1: Does the user have a Microsoft 365 Copilot license?

If yes: core Copilot Chat, Microsoft 365 app experiences, native Work IQ grounding, Researcher, Analyst, Facilitator, and qualifying employee-facing agents are generally included. Then inspect the workload for explicit exceptions.

Standard or Copilot Chat harness, authenticated B2E use: included for qualifying execution in Microsoft 365 channels, subject to service and fair-use limits.
GitHub Copilot harness in Copilot Studio: Copilot Credits are required for its metered build, test, evaluation, and runtime activities. This is the Copilot Studio agent harness, not the separately licensed GitHub Copilot developer product.
Computer-using agent, autonomous execution, external or unauthenticated channel, or BYOM: credits and/or a separate service charge can apply.
Copilot Cowork: the Copilot license is required, but Cowork consumption is not included.
Work IQ API: API use is metered even though native Work IQ grounding inside the licensed Copilot experience is included.

If no Copilot add-on, but the user has an eligible Microsoft 365 or Office 365 license: Copilot Chat Basic, web grounding, uploads, supported app context, and lightweight instruction or public-website agents can be available without Copilot Credits. Tenant-grounded agents, SharePoint agents, and Copilot Studio capabilities can require organizational usage-based billing.

Microsoft Copilot Credits decision map showing included experiences, metered Copilot Credit paths, separate billing models, purchasing options, standard-harness event examples, and a callout explaining that GitHub Copilot is licensed and metered separately — **Figure 1.** Decision map for identifying when Microsoft Copilot usage is subscription-included, Copilot Credit-metered, unavailable without a prerequisite, or governed by a separate capacity model. The decisive factors are the user license, experience, agent harness, identity and context, channel, and workload. The footer callout separates the GitHub Copilot developer product from the GitHub Copilot harness inside Copilot Studio.

GitHub Copilot is a separate product, a separate license, and a separate meterGitHub Copilot is licensed per developer seat under its own plans, and GitHub Copilot Chat in the IDE and on GitHub.com is included with a paid seat. None of that consumes Microsoft Copilot Credits, and a Microsoft 365 Copilot subscription does not grant it. GitHub bills on its own meter: since June 1, 2026, GitHub Copilot usage is measured in GitHub AI Credits, where each plan carries a monthly allowance, chat, agent mode, code review, and CLI activity draw from it, and code completions and next edit suggestions remain unlimited on paid plans. The GitHub Copilot harness named in the decision map is something different: it is the agent-building harness inside Copilot Studio, and that harness is what consumes Copilot Credits. Two products, two contracts, two budgets.

The fastest diagnostic questionDo not begin with “Does the organization own Copilot?” Begin with “Who is executing which experience, under what identity, in which harness and channel, against which data, and what tools or triggers are invoked?”

Detailed ecosystem matrix

Experience	User or scenario	Answer	Billing trigger and important condition
Microsoft Copilot core experience	Microsoft 365 Copilot-licensed user	🟩 Included	Core chat, supported app experiences, Microsoft-built agents, native Work IQ grounding, and model access are included. Service limits can still apply.
Copilot Chat Basic	Eligible user without Copilot add-on	🟩 Included	Web grounding, uploaded content, image generation, and supported app context use the baseline service. Broad work grounding is not equivalent to the licensed experience.
Declarative agent using public content	Licensed or unlicensed Copilot Chat user	🟩 Included	Instructions and public websites only, without shared tenant-data retrieval or metered capabilities.
Tenant-grounded Copilot Chat agent	User without Copilot add-on	🟨 Copilot Credits	SharePoint, Graph or Copilot connectors, and shared tenant data cause metered consumption. Billing must be enabled.
SharePoint agent	Licensed Copilot user	🟩 Included	Agent use is included, but the user still requires permission to the underlying content.
SharePoint agent	User without Copilot add-on	🟨 Copilot Credits	Organization-enabled SharePoint agent usage consumes credits under its supported billing configuration.
Copilot Studio Standard or Copilot Chat harness	Licensed, authenticated employee-facing use	🟩 Included	Qualifying answers, actions, tenant Graph grounding, and properly invoked agent flows are zero-rated when the agent acts as the licensed user.
Copilot Studio Standard harness	Unlicensed, anonymous, external, or otherwise nonincluded use	🟨 Copilot Credits	Feature-based rates apply, and one interaction can invoke multiple metered events.
Autonomous or scheduled execution	No qualifying licensed-user context	🟨 Copilot Credits	Standard rates apply. Background automation is not made universally free by owning a Copilot seat.
Computer-using agent	Including licensed-user initiation	🟨 Copilot Credits	Computer use is billed as an agent action and can have wider Power Automate or RPA licensing dependencies.
GitHub Copilot (developer product)	Developer with a GitHub Copilot seat	🟥 Separate meter	Requires its own GitHub Copilot seat, which includes GitHub Copilot Chat in supported IDEs and on GitHub.com. Usage is measured in GitHub AI Credits, not Copilot Credits, and code completions remain unlimited on paid plans.
GitHub Copilot harness in Copilot Studio	Maker and runtime scenarios	🟨 Copilot Credits	LLM-assisted building, knowledge, tools, evaluation, and runtime are usage-based. Do not assume the standard B2E zero-rating. This is the Copilot Studio harness, not the GitHub Copilot seat.
Copilot Cowork	Microsoft 365 Copilot-licensed user	🟨 Copilot Credits	Requires a seat plus enabled usage billing and a spending policy. Tasks consume credits based on work performed.
Work IQ in licensed Copilot	Normal Copilot interaction	🟩 Included	Native grounding across organizational content is included. Direct API invocation is a different entitlement.
Work IQ APIs	Agent or application	🟨 Copilot Credits	Variable query usage applies; the Work IQ Tools API is listed at 0.1 credit per API call.
Dynamics 365 and Power Platform AI	Product-dependent agent or AI feature	🟦 Varies	Credits and included amounts depend on the product entitlement and AI capability invoked. Other licensing can remain necessary.
Security Copilot	Security portal, embedded experience, or security agent	🟥 Separate meter	Uses Security Compute Unit capacity, not Copilot Credits.
Azure AI Foundry or BYOM	Model hosted outside Copilot Studio	🟥 Separate meter	Azure model, token, and hosting charges are separate. Copilot Studio can still meter an action that invokes the tool.

Copilot Studio: the conditions that matter

A Microsoft 365 Copilot license can cover employee-facing Copilot Studio use when the consuming user is licensed and authenticated, the agent acts as that user, the use case is business-to-employee, the execution occurs through a qualifying Microsoft 365 channel, and an agent flow uses the qualifying When an agent calls the flow invocation pattern.

Common exclusionsCustomer-facing or anonymous use, external users, no-authentication agents, autonomous triggers, agent flows invoked by another trigger, computer-using agents, the GitHub Copilot harness, separately hosted models, and downstream licensing requirements can fall outside the licensed-user inclusion.

Standard-harness published-use rate card

Published Standard-harness rates use several different denominators, so they cannot honestly be ranked against one another on a single axis. The three charts below are therefore split by unit. Each uses a linear scale, and each is scaled to the largest value in its own chart.

Per-event rates

Copilot Credits per response, action, message, or page. Agent-flow actions are published at 13 credits per 100 actions and are shown here divided down to a per-action figure so they can be compared fairly.

Agent flow action0.13 credits per action, published as 13 per 100

Classic answer1 credit per response

Generative answer2 credits per response

Agent action5 credits per action

Content processing8 credits per page or image

Tenant Graph grounding10 credits per message

Text and generative AI tools

Copilot Credits per 10 responses. Microsoft also publishes these as 0.1, 1.5, and 10 credits per 1,000 tokens, per 1,000 characters, per image, or per page. This is the tier that most often turns a cheap agent into an expensive one.

Basic1 credit per 10 responses

Standard15 credits per 10 responses

Premium100 credits per 10 responses

Voice tiers

Copilot Credits per minute. Core agent activity such as classic answers, generative answers, and agent actions is included within the voice rate.

Basic, classic orchestration10 credits per minute

Standard, generative orchestration35 credits per minute

Premium, real time75 credits per minute

Meter stacking mattersA single response can invoke more than one meter. A generative answer grounded in tenant Graph is the simple example: 2 credits for the answer plus 10 credits for tenant Graph grounding, totaling 12 credits.

Reasoning models bill on two meters at onceA reasoning-capable model is charged the normal feature rate for the operation plus the premium text and generative AI tools rate for the reasoning itself, at 10 Copilot Credits per 1,000 tokens. This is why the same agent design can cost 2 credits per answer on a standard model and well over 100 once tenant grounding and a reasoning model are switched on. Feature creep on an already-published agent is the most common cause of a consumption surprise, because no new agent is approved and no new project is opened.

Authoring and testing are not all treated the same

For the Standard harness, billing generally begins after publication, while test-panel prompts are free. However, prompts or models inside agent flows can consume credits during testing. In the GitHub Copilot harness, natural-language building, previewing, testing, evaluation, and runtime are usage-based; manual non-LLM configuration is treated differently.

Copilot Cowork: access is licensed, work is metered

Copilot Cowork for work and school accounts became generally available on June 16, 2026. Cowork requires an assigned Microsoft 365 Copilot user subscription, tenant enablement, usage-based billing, an applicable spending policy, and supported tenant and client prerequisites.

The subscription does not create a Cowork credit entitlement. Consumption varies with model choice, context retrieval, tool and skill calls, orchestration and runtime, image generation, and browser tasks. Microsoft does not publish a single fixed credits-per-prompt rate. The /cost command can show an approximate aggregate for a task, but it is not the official invoice ledger.

Buying Copilot Credits

Purchasing route	Commercial structure	Best fit
PAYG	$0.01 per credit, billed in arrears through the connected Azure billing arrangement	Pilots, uncertain demand, or seasonal usage
Copilot Studio capacity pack	$200 per month, billed annually; 25,000 credits per month; unused monthly credits do not roll over	Stable monthly consumption
Copilot Credit Pre-Purchase Plan (P3)	One-year upfront pool with published volume tiers	Predictable enterprise-scale usage
Microsoft Agent P3	Broader annual commitment for eligible agentic usage across supported platforms	Cross-platform agent estates

Supported purchasing options can coexist. Prepaid capacity can cover planned consumption while PAYG provides an overage path.

Simple list-price comparison

Monthly list cost

US list price per month, on a linear scale.

PAYG at 10,000 credits$100

PAYG at 20,000 credits$200

One 25,000-credit pack$200

PAYG at 25,000 credits$250

Break-even at list price: $200 ÷ $0.01 = 20,000 credits per month

Below 20,000 credits, PAYG has the lower simple list cost. Above 20,000 credits, a fully utilized 25,000-credit pack has the lower list cost, subject to the annual billing commitment and the loss of unused monthly credits.

Illustrative workload calculations

Graph-grounded generative agent: 200 unlicensed users × 20 interactions per month × 12 credits = 48,000 credits, or $480 PAYG. Two capacity packs provide 50,000 credits for $400 per month under the applicable annual commitment.
Cowork planning example: if observed usage averages 300 credits per task, 50 users × 10 tasks per month × 300 = 150,000 credits, or $1,500 PAYG. Six capacity packs provide the same 150,000 credits for $1,200 per month under the applicable annual commitment. The 300-credit value is an illustrative planning assumption, not a Microsoft fixed rate.

Do not turn the illustration into a quoteActual pricing can vary by CSP margin, EA or MACC arrangement, currency, geography, tax, contract date, and negotiated discounts. Use observed tenant consumption and the customer’s purchasing agreement for the final model.

Terminology and traps

“Messages” is legacy terminology. The common agent consumption currency moved to Copilot Credits, but older documentation and operational language can still surface.
Microsoft Copilot and Microsoft 365 Copilot can coexist in documentation. Confirm whether a page is discussing the product family, the paid user entitlement, or a usage-based service.
GitHub Copilot is not Microsoft 365 Copilot, and GitHub AI Credits are not Copilot Credits. Developer seats are bought, licensed, and metered separately, so they do not belong in a Copilot Credit consumption model. The only overlap in this guide is the similarly named Copilot Studio harness.
Copilot Credits are not consumer AI credits. Consumer Microsoft 365 subscription limits are a different construct and should not be used for tenant planning.
A licensed user does not create a shared credit pool for coworkers. The license covers that person’s qualifying use; other metered users need organizational funding.
Not every agent interaction is charged. Public-content declarative agents and qualifying authenticated employee-facing use can be included.
Credits do not replace other licensing. Dataverse, Power Platform premium rights, Dynamics restrictions, external-user access, Azure-hosted models, and automation dependencies can still apply.

Recommended client governance baseline

Segment the audience into licensed frequent Copilot users, unlicensed Copilot Chat users, Cowork users, agent makers, external or customer audiences, and developers on GitHub Copilot seats, which are budgeted and metered outside this model.
Inventory every agent’s harness, authentication, channel, data sources, triggers, tools, owning environment, and downstream services.
Begin uncertain workloads with PAYG and explicit user or group controls, then consider packs or P3 after observing stable consumption.
Treat a Cowork spending policy as both cost control and access control. A policy that permits consumption also enables access until the configured limit is reached.
Monitor Cowork and Work IQ usage in Microsoft 365 Cost Management and Copilot Studio consumption in the Power Platform admin center, recognizing that operational views differ by workload.
Revalidate the decision map before a large order because licensing guides, harness availability, pricing, and preview boundaries can change.

Bottom lineA Microsoft 365 Copilot subscription covers everyday licensed-user productivity. Copilot Credits fund separately metered work. The boundary is determined by the user, identity, harness, channel, data, trigger, and capability, not by the word “Copilot” alone.

Blog