Billing rates and management

This article describes the rates for the different features and capabilities used in agents that Copilot Studio bills through the pay-as-you-go meter or Copilot Credit packs. These rates apply to all language models that Copilot Studio provides. They exclude bring-your-own-model configurations, including Azure Foundry models, which are billed separately.

Copilot Credits are the unit that measures agent usage. You calculate the total cost based on the sum of the Copilot Credits your organization uses. The number of Copilot Credits an agent consumes depends on the design of the agent, how often customers interact with it, and the features they use.

When you purchase a Copilot Studio license, you get a specific number of billed Copilot Credits. You pool this capacity across the entire tenant.

Copilot Credits billing rates

The following table shows the different billing rates. Each interaction with an agent might use multiple feature types at the same time. For example, an agent grounded in a tenant graph could use 12 Copilot Credits (10 Copilot Credits for tenant graph grounding, and 2 Copilot Credits for generative answers) to respond to a single complex prompt from a user.

¹ 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 Copilot USL when the user of the agent is licensed with Microsoft Copilot, and the agent operates using the authenticated Microsoft Copilot USL user's identity. For agent flows, the "No charge" inclusion applies only to runs triggered via the "When an agent calls the flow" trigger, where the user interacting with the agent is authenticated with a Microsoft 365 Copilot license. Agent flows using other triggers consume Copilot Credits at the standard rate. For these flows, all actions in the agent flow, including calls to agents, consume credits at the standard rate. Usage is limited to fair usage limits. Microsoft reserves the right to update the limits as the product evolves, and new usage patterns emerge.

² Agents incur charges for generative answer responses unless the agent is created in Agent Builder in Microsoft 365 and the response doesn't leverage tenant graph grounding.

³ When the environment has no Copilot Credits available, agent flow enforcement blocks new agent flow runs. It doesn't block the agent that calls the flow. Footnote 1 describes when the Microsoft 365 Copilot inclusion applies to agent flows.

⁴ Computer-Using Agents (CUA) usage is not included in the Microsoft Copilot USL.

• Classic answers: These events are predefined responses manually authored by agent makers. They're static and don't change unless manually updated. Use them where you want precise and controlled responses that are the only ones the agent generates.

• Generative answers: These events are dynamically generated using AI models, such as Generative Pretrained Transformers (GPTs). They can adapt and change based on the context and the knowledge sources they're connected to. They're useful for handling a wide range of topics and providing more flexible and natural interactions.

• Tenant graph grounding for Copilot Credits: These events provide higher quality grounding for your agents by using retrieval-augmented generation (RAG) over your tenant-wide Microsoft Graph, including external data synced into Microsoft Graph through connectors. This capability results in more relevant and improved responses and ensures that the grounding information is up-to-date. This capability is optional, and you can turn it on or off for each agent. Learn more in Tenant graph grounding with semantic search.

• Agent actions: Agent actions refer to steps such as triggers, deep reasoning, and topic transitions that appear on the activity map in Copilot Studio when testing an agent. In addition, Computer-Using Agents are also billed at the agent action rate.

• Text and generative AI tools: Prompt tools embedded within an agent enable the creator to direct the underlying model to perform intelligent document and image processing tasks, behave in a task-specific manner, or generate scenario-specific outputs. The three types of tools (basic, standard, and premium) are based on the underlying language model of the prompts. The premium text and generative AI tools item charges for advanced reasoning in agents. Learn more in Licensing and Copilot Credits and Prompt tokens in the AI Builder documentation.

• Agent flow actions: This item charges for agent flows that enhance AI agents with agent flows, which are predefined sequences of flow actions to execute repetitive tasks quickly, without requiring agent reasoning and orchestration at each step. Learn more in Agent flows overview.

⁵ When a reasoning model orchestrates your agent, this charge is calculated from the exact number of billed tokens, not from the number of responses. One response can involve many thousands of tokens. How tokens are counted for reasoning models describes which tokens are billed. Prompt tools are measured in the units described in Licensing and Copilot Credits in the AI Builder documentation.

Reasoning model billing rates

When an agent uses a reasoning-capable language model, Copilot Studio bills by using two billing meters: feature rate and text and generative AI tools (premium).

The feature rate covers the core action the agent performs, such as generative answers, executing actions in flows, or performing agent actions. This rate always applies, regardless of the model used.

The text and generative AI tools (premium) rate applies when the agent uses a reasoning model. This premium rate covers the extra computational resources required for deep reasoning, planning, and multistep inference. It's billed as Text and generative AI tools (premium) per 1000 tokens, at 10 Copilot Credits.

The cost calculation for an operation that uses a reasoning model is as follows:

Total cost = feature rate for the operation + text and generative AI tools (premium) for the reasoning model's token usage.

For example, a generative answer that uses a reasoning model is billed as the following formula:

Total cost = feature rate for generative answers + text and generative AI tools (premium) per 1000 tokens.

How tokens are counted for reasoning models

When a reasoning model orchestrates your agent, including any child agents, the text and generative AI tools (premium) charge is based on the tokens that the model processes, except the tokens that it reads from its prompt cache, as the following table shows. The charge isn't based on the number of messages. To respond to one message, the agent can call the model several times, for example to plan, to search knowledge or call tools, and to write the response. Each call counts.

Each model call includes:

• Your agent's instructions and the descriptions of its tools and knowledge sources.
• The conversation so far.
• The knowledge and tool results that the agent retrieved during the turn.

The following table shows which tokens are billed.

A prompt cache stores the beginning of a prompt, such as the instructions and the earlier part of the conversation, so the model doesn't have to process it again on the next call. Tokens that the model reads from its cache aren't billed. Content that the model processes for the first time, such as a new question or new knowledge results, is written to the cache and billed. Content that's the same for every conversation, such as instructions, might already be in the cache when a conversation starts. Cached content is kept for a limited time. If the model has to process the content again after it expires, those tokens are billed again.

Copilot Credits are calculated from the exact number of billed tokens, at 10 Copilot Credits per 1,000 tokens.

A few messages can therefore consume many Copilot Credits when your agent has long instructions, retrieves a lot of knowledge, or holds long conversations. To reduce consumption:

• Use a model that isn't a reasoning model when your scenario doesn't need deep reasoning.
• Keep instructions concise and knowledge sources focused.
• Start a new conversation for a new topic instead of resuming a long conversation.
• Ask your admin to set a monthly limit for the agent. Learn more in Manage Copilot Credits and capacity for Copilot Studio.

Agents powered by the GitHub Copilot harness are billed differently. Learn more in Overview of usage-based billing.

Billing examples

The following billing examples show how an agent might incur billing, based on its configuration.

Customer support agent

You have a customer support agent on your website that answers questions based on customer return policies and product manuals that you provided to the agent as a knowledge source.

An average run includes four classic answers for return-related questions and two generative answers for troubleshooting questions. The average is 900 customers per day. The estimated cost per day is based on the following calculation: [(4x1)+(2x2)] x 900 customers = 7200 Copilot Credits.

Sales performance agent

You have a tenant graph grounded agent in Microsoft Copilot Chat. This agent answers employee questions based on sales data connected to Microsoft Graph using Graph data connectors.

An average run comprises four generative answers and four tenant graph grounded Copilot Credits. The average is 50 Microsoft 365 Copilot licensed users and 100 unlicensed users. The estimated cost per day is based on the following calculation: [(4x2)+(4x10)] x 100 users = 4,800 Copilot Credits.

Order processing agent

An internal-facing agent is autonomously triggered anytime a new order is received by the organization. The agent uses a single knowledge source to get product details about items ordered, and triggers 4 action calls to confirm product availability, view shipping timelines, approve the order, and send an email to the customer with all details. Actions and topics are agent actions in generative orchestration mode. The estimated cost per day is based on the following calculation: [(4x5)] = 20 Copilot Credits.

Customer support agent that uses a reasoning model

A customer support agent on your website uses Claude Opus 4.7, a reasoning model, to orchestrate its responses. It answers questions from product manuals that you provided as a knowledge source. A customer asks a follow-up question in an ongoing conversation. To answer it, the agent calls the model twice:

1. Plan: The model processes 42,000 input tokens—the agent's instructions, the descriptions of its tools and knowledge sources, the conversation so far, and the new question. It reads 38,000 of these tokens from its prompt cache and writes the other 4,000 tokens to the cache. It returns 600 output tokens, including its reasoning and a request to search the product manuals.

2. Answer: The model processes the same 42,000 tokens, now read from the cache, plus the 600 tokens from the first call and 8,000 tokens of search results. It writes these 8,600 new tokens to the cache and returns a 900-token answer.

The Text and generative AI tools (premium) charge for this message is calculated as follows: 14,100 billed tokens x 10 Copilot Credits / 1,000 tokens = 141 Copilot Credits. The feature rates for the activities in the turn are added to this amount.

If the same question starts a new conversation, content that's the same for every conversation, such as instructions, could already be in the cache. For example, if the plan call read 6,000 tokens from the cache and wrote the other 36,000 input tokens to it, that call would bill 36,600 tokens, and the message would cost: (36,600 + 9,500) x 10 / 1,000 = 461 Copilot Credits.

The numbers in this example are illustrative. The token counts for your agent depend on its instructions, knowledge sources, tools, and conversations.

Overage enforcement

When consumption exceeds available capacity, the environment is in overage. Microsoft allows some level of overage consumption, similar to a grace period, to avoid blocking business processes.

If your environment has no more capacity, you have the following options:

• Reallocate existing capacity from the organization (tenant) or environment level.

• Purchase more capacity and make it available to your environment.

• Set up a consumptive meter or pay-as-you-go meter to handle the overage.

Enforcement policy

Applies to all tenants operating under the Copilot Studio prepaid capacity model for custom agent usage (conversational and autonomously triggered).

Usage threshold

Enforcement is triggered when a tenant reaches 125% of their prepaid capacity.

Action on overage (125%)

When enforcement applies, the environment's agents are blocked. A blocked agent stops responding, including in a conversation that's already in progress.

If an environment bills overage to a pay-as-you-go plan, running out of prepaid Copilot Credits doesn't block the agents in that environment, and usage that prepaid capacity doesn't cover is billed to the plan. Agents that an admin turned off, or that reached a monthly limit set to stop usage, are still blocked. Learn more in Usage limit and agent unavailable errors.

Agent flow runs are blocked separately from agents. New agent flow runs can be blocked while the agent keeps answering questions that don't need a flow.

Notification mechanism

The tenant's designated administrator receives an email notification, and the Power Platform admin center also displays the notification.

Agent behavior after enforcement

When enforcement triggers, most users see this message: "This agent is currently unavailable. It has reached its usage limit. Please try again later."

Users who sign in with a Microsoft 365 Copilot license might not be blocked.

Enforcement example

If you allocate or reserve capacity in an environment, the system honors that capacity. Consider the following example of a customer with four different environments, and how their Copilot Credit capacity is enforced.

A customer has 25,000 Copilot Credits, and uses the following allocation structure:

• Environment A has 10,000 Copilot Credits allocated.
• Environment B has no allocation.
• Environment C has no allocation.
• Environment D has an allocation of 500 Copilot Credits and uses pay-as-you-go.

Environment B and Environment C have no allocation and draw from the tenant's capacity.

If Environment A draws or consumes Copilot Credits against its allocation of 10,000 Copilot Credits, the following scenario applies. When the 10,000 Copilot Credits are consumed, Environment A can consume from the tenant.

If Environment A consumes from the tenant, it joins Environment B and Environment C in consuming from the tenant capacity. If the tenant reaches 125% Copilot Credit consumption, enforcement is invoked.

If the tenant is already at 125% Copilot Credit consumption because of Environment B and Environment C, enforcement isn't placed on the agents in Environment A, as long as Environment A has remaining capacity from its allocation of 10,000 Copilot Credits.

For Environment D, when the tenant is in overage, this environment isn't impacted. Because once Environment D reaches its 500 Copilot Credit limit, the pay-as-you-go meter is invoked.

Agent flow enforcement

Agent flow enforcement applies specifically to agent flows and not Power Automate cloud flows. It works differently from general agent enforcement.

When an environment has no Copilot Credits available, new agent flow runs are blocked. An environment has credits available when it has allocated prepaid credits left, when it can draw from unallocated prepaid capacity in the tenant, or when it's linked to a pay-as-you-go billing plan that includes Copilot Studio. Unlike the enforcement that blocks agents, agent flow enforcement blocks only new agent flow runs. The parent agent continues to function normally for all non-flow interactions, such as classic answers, generative answers, and agent actions.

Enforcement behavior

When agent flow enforcement is active:

• New agent flow runs can't start. Attempts to trigger an agent flow are rejected.
• Agent flow runs that are already in progress complete normally.
• The agent remains available for non-flow interactions.
• Flow authors see a design-time warning in the Copilot Studio designer indicating that capacity is exhausted.
• Enforcement resets monthly when prepaid Copilot Credits are renewed.

Exemptions

Agent flow enforcement doesn't apply to the following scenarios:

• Microsoft 365 Copilot licensed users: When a user who is authenticated with a Microsoft 365 Copilot license interacts with an agent, and the agent calls an agent flow that uses the When an agent calls the flow trigger, the agent flow actions are included at no charge. These runs aren't charged and aren't subject to enforcement.

  This exemption doesn't apply when the agent runs autonomously, for example after an event trigger or a Power Automate flow starts it. Agent flow runs called by an autonomous agent consume Copilot Credits and are subject to enforcement, even if the users in your organization have Microsoft 365 Copilot licenses.

• Test runs: Agent flow runs that the agent starts from the test chat in Copilot Studio and test runs that you start in the flow designer, don't consume capacity for agent flow actions and aren't blocked by enforcement.

These exemptions apply to agent flow actions. AI tools that run inside an agent flow, such as prompts, are metered separately. Learn more in Licensing and Copilot Credits.

Resolve agent flow enforcement

If agent flow enforcement is active in your environment, you have the following options:

• Reallocate existing capacity from other environments in your tenant. In the Power Platform admin center, go to Licensing > Copilot Studio and select Manage Copilot credits.
• Purchase more Copilot Credits and assign them to the affected environment.
• Enable pay-as-you-go billing by linking an Azure subscription billing plan that includes Copilot Studio to your environment. With pay-as-you-go, enforcement doesn't apply because overage is billed to your Azure subscription. Learn more in Copilot Studio usage limits and agent unavailable errors.

To monitor agent flow credit consumption, go to Licensing > Copilot Studio > Environments, select the affected environment, and review the Agent flow actions line in the Copilot credit consumption details grid. Learn more in Manage Copilot Credits and capacity for Copilot Studio.

View Copilot Credit consumption

You can view Copilot Credit consumption reporting in the Power Platform admin center.