This article is part of our AI Agents for Business hub. Want help applying it? Talk to eSEOspace.
Copilot Studio is billed in Copilot Credits, sold in packs of 25,000 for $200 per month. Your real cost depends on what each agent does. Microsoft's billing rates range from 1 credit for a scripted answer to token-based charges for reasoning models.
This guide explains the meter in plain terms, and where bills tend to surprise people. It is part of our AI agents for business hub. Prices and rules were checked against Microsoft's documentation on 9 November 2026, and they change often.
Key Takeaways
- Microsoft sells tenant-wide packs of 25,000 Copilot Credits for $200 per month, or usage through pay-as-you-go.
- A generative answer costs 2 credits, an agent action 5, and tenant graph grounding 10.
- Reasoning models add 10 credits per 1,000 billed tokens, on top of the feature rate.
- Prepaid agents are blocked at 125% of capacity, including during a live conversation.
- Microsoft 365 Copilot users are not charged for some usage, but autonomous runs and computer-using agents are excluded.
The rates
Microsoft's billing rates page lists what each feature costs for agents on the standard harness. Microsoft says agents on the GitHub Copilot harness are billed differently. One interaction can use several at once, so a single reply may add up to more than one line.
| Feature | Copilot Credits |
|---|---|
| Classic answer | 1 |
| Generative answer | 2 |
| Agent action | 5 |
| Tenant graph grounding | 10 |
| Agent flow actions | 13 per 100 actions |
| Text and generative AI tools, basic | 1 per 10 responses |
| Text and generative AI tools, standard | 15 per 10 responses |
| Text and generative AI tools, premium | 10 per 1,000 tokens |
| Content processing tools | 8 per page |
Microsoft gives a simple example. A website support agent averages four classic answers and two generative answers per customer, with 900 customers a day. That comes to 7,200 credits a day.
In a 30-day month, that is 216,000 credits, or about nine packs at $1,800. That arithmetic is ours, and it assumes every month looks the same.
How you pay
Microsoft sells credit packs at $200 per pack per month. At that rate, one credit works out to $0.008 by our math. Packs are pooled across the whole tenant.
There are three ways to buy, according to Microsoft's licensing page.
| Option | How it works | Note |
|---|---|---|
| Prepaid packs | Monthly capacity, pooled per tenant | Unused credits do not carry over |
| Pay-as-you-go | Billed monthly through Azure | No upfront commitment |
| Prepurchase plan | One-year pool of Commit Units | Microsoft's product page says up to 20% savings |
We could not confirm the pay-as-you-go rate per credit. Microsoft's product page and Azure pricing page did not display a figure when we checked. Get it from Azure's calculator or your reseller before budgeting.
On rollover, Microsoft's licensing page says unused credits do not carry over to the next month. Its estimator page is looser, and says to check your agreement. Confirm what your contract says.
Reasoning models change the math
This is where budgets go wrong. Microsoft says a reasoning model adds 10 credits per 1,000 billed tokens, on top of the normal feature rate. It charges on tokens, not messages, so one response can involve many thousands of tokens.
Each model call counts. The agent's instructions, the conversation so far and any retrieved knowledge are all part of it. Output tokens are billed too, including the tokens the model uses to reason.
Microsoft publishes an illustrative example, which it says is not a forecast for your agent.
| Scenario | Billed tokens | Credits | At the pack rate |
|---|---|---|---|
| Plain generative answer | Not token-billed | 2 | About $0.016 |
| One follow-up message | 14,100 | 141 | About $1.13 |
| Same question, new conversation | 46,100 | 461 | About $3.69 |
The dollar column is our arithmetic. The two token rows show the reasoning charge only, so the feature rate is extra. A new conversation costs more in this example because less is read from the prompt cache.
Microsoft advises using a non-reasoning model when you do not need deep reasoning. It also suggests short instructions, focused knowledge sources and a new conversation for each new topic. Our AI agent pricing guide and cost per task article look at this across vendors.
What "no charge" really covers
The billing table shows "No charge" for users with a Microsoft 365 Copilot license. It is narrower than it sounds. Microsoft says the inclusion applies to employee-facing use when the agent operates under that licensed user's identity.
These cases still consume credits or are not covered:
- Customer-facing agents, where users do not hold the license.
- Autonomous runs, such as an agent started by an event trigger.
- Agent flows on any trigger other than "When an agent calls the flow."
- Computer-using agents.
- Usage beyond Microsoft's fair use limits, which Microsoft can update.
The two Microsoft pages differ slightly. The billing table marks premium reasoning as no charge, while the licensing page lists classic answers, generative answers and tenant grounding. Ask Microsoft in writing if your plan depends on reasoning usage being free.
The 125% cutoff
If you use prepaid capacity only, you have a hard ceiling. Microsoft says enforcement is triggered when a tenant reaches 125% of its prepaid capacity. For a single 25,000-credit pack, that is 31,250 credits by our math.
At that point the environment's agents are blocked. A blocked agent stops responding, including in a conversation already in progress. Most users see a message saying the agent has reached its usage limit.
A few details change how this plays out:
- An environment billing overage to pay-as-you-go is not blocked when prepaid credits run out.
- Environments with their own allocation are treated separately from the shared tenant pool.
- Agent flows are blocked separately, when the environment has no credits available.
- Users with a Microsoft 365 Copilot license might not be blocked.
How to keep the bill under control
Microsoft provides the tools, but they need setting up.
- Set a monthly limit per agent in the Power Platform admin center, with alerts and an optional hard stop.
- Remember that limits are not instant, and an agent's limit does not stop the flows it calls.
- Download the consumption reports. They show use by environment, agent and user, but not per conversation or token counts.
- Use Microsoft's estimator for planning, but add the 10 to 20% buffer it recommends.
If you plan to connect agents to outside tools, our Copilot Studio MCP setup guide covers that side.
Next step
If you want help estimating the cost of an agent before you commit, see our AI agent development service. We can review your planned agent and identify which steps are likely to drive credits.
Frequently asked questions
How much does Copilot Studio cost?
Microsoft sells packs of 25,000 Copilot Credits for $200 per month, or you can pay as you go through Azure. Your actual cost depends on which features each agent uses and how often.
What is a Copilot Credit?
Copilot Credits are the unit Microsoft uses to measure agent usage. A classic answer costs 1 credit and a generative answer costs 2, while reasoning models are charged per 1,000 tokens.
Is Copilot Studio included with Microsoft 365 Copilot?
Partly. Employee-facing use under a licensed user's identity shows no charge, within fair use limits. Autonomous runs, computer-using agents and some flows still consume credits.
What happens when I run out of credits?
If you use prepaid capacity only, agents are blocked once the tenant reaches 125% of capacity. An environment linked to pay-as-you-go continues and is billed for the extra usage.
Do unused Copilot Credits roll over?
Microsoft's licensing page says unused credits do not carry over to the next month. Its estimator page says to check your specific agreement, so confirm with your contract.
