Contents
Fast answer
Copilot Studio is sold as tenant-wide prepaid packs of 25,000 Copilot Credits at $200 a pack per month, which works out at $0.008 a credit, with pay-as-you-go through a linked Azure subscription as the alternative. Microsoft publishes the credit rate card: a classic answer costs 1 credit, a generative answer 2, an agent action 5, tenant graph grounding 10, agent flow actions 13 per 100 actions, and content processing 8 credits a page. Text and generative AI tools cost 1, 15 or 100 credits per 10 responses on the basic, standard and premium tiers, and reasoning models bill the premium tool rate of 10 credits per 1,000 tokens on top of the feature rate. Voice is the expensive meter: 10 credits a minute for classic voice, 35 for GenAI voice and 75 for premium GenAI voice, or roughly $0.08, $0.28 and $0.60 a minute at the pack rate. Unused credits do not carry over, and at 125 percent of prepaid capacity Microsoft disables custom agents.
Facts last verified 2026-08
Quick facts
- Prepaid pack
- $200.00 per pack per month for 25,000 Copilot Credits (checked 2026-08-20)
- Effective credit price
- About $0.008 per Copilot Credit
- Pay-as-you-go
- Billed monthly to a linked Azure subscription for credits used
- Generative answer
- 2 credits, about $0.016
- Agent action
- 5 credits, about $0.04
- Tenant graph grounding
- 10 credits per message, about $0.08
- GenAI voice
- 35 credits a minute, about $0.28
- Enforcement
- Custom agents disabled at 125 percent of prepaid capacity
On this page
What this page covers
A comparison visitor should understand the tradeoff, the best-fit scenario, and the next diagnostic tool to confirm the choice.
- 01What it costs
- 02The credit rate card
- 03Worked example
- 04Where the meter bites
- 05Cheapest honest path
- 06When to leave
Why does this matter now?
Copilot Studio is the only platform in this set that prices in an abstract currency, and that abstraction is where budgets go wrong. Nobody intuitively knows what 25,000 credits buys, so teams launch, watch consumption, and find out. Converting the published rate card into dollars before you build turns a vague commitment into arithmetic you can defend, and it exposes the two genuine cliffs: voice minutes at 35 or 75 credits, and the enforcement threshold that switches your agents off in front of customers.
Internal path
Where to go next from this page
These links are part of the A8gent learning and conversion path. Use them to move from concept, to diagnosis, to workflow build, to course.
What you should be able to do after this
- Convert Copilot Credits into dollars
- Price a conversation before you build it
- See why voice costs so much more
- Avoid the overage enforcement cliff
How do you do it, step by step?
1. Convert credits to dollars first
A pack is 25,000 Copilot Credits for $200 a month, so a credit is $0.008. Every rate on Microsoft's table becomes money at that multiplier: a generative answer is $0.016, an agent action is $0.04, tenant graph grounding is $0.08 a message, and content processing is $0.064 a page. Do this once and write it on the wall.
2. Cost one conversation, then multiply
Microsoft's own worked example is a support agent averaging four classic answers and two generative answers per run, which is 8 credits a conversation. At 900 conversations a day that is 7,200 credits a day. Build the same sum for your agent from its actual design, because an agent grounded in the tenant graph costs 12 credits for a single complex prompt rather than 2.
3. Price voice separately and carefully
GenAI voice is 35 credits a minute and premium GenAI voice is 75, which at the pack rate is about $0.28 and $0.60 a minute. Core agent activity is included in those voice rates, but they are still several times what a specialist voice platform charges per minute. If voice is the main use case, price a dedicated voice platform against this before committing.
4. Watch for the reasoning model double charge
When an agent uses a reasoning-capable model, Copilot Studio bills the feature rate plus text and generative AI tools at the premium rate of 10 credits per 1,000 tokens. A generative answer that would have cost 2 credits can therefore cost many multiples of that. Turn reasoning on deliberately, per topic, rather than as a default.
5. Set consumption limits before you hit enforcement
Enforcement triggers at 125 percent of prepaid capacity, at which point custom agents are disabled and end users see a billing issue message. Unused credits do not carry over month to month, so both over-buying and under-buying are penalised. Microsoft lets you cap credit usage per agent in the Power Platform admin centre, and linking a pay-as-you-go meter means overage bills instead of blocking.
Converting an abstract currency into a budget line
Copilot Studio is the only platform in this batch that prices in a currency it invented, and that abstraction is where budgets go wrong. Nobody has an intuition for what 25,000 Copilot Credits buys, so teams launch, watch consumption for a month, and find out empirically. The conversion is trivial once you do it: a prepaid pack is 25,000 credits at $200.00 a month, so a credit is $0.008. Write that number down, because every rate Microsoft publishes becomes money the moment you multiply by it, and none of them mean anything before.
Applying it to the published rate card gives you a price list you can actually reason about. A classic answer at 1 credit is $0.008. A generative answer at 2 credits is $0.016. An agent action at 5 credits is $0.04. Tenant graph grounding at 10 credits is $0.08 a message. Agent flow actions at 13 credits per 100 actions is about $0.104 per hundred. Content processing at 8 credits a page is $0.064. Text and generative AI tools run 1, 15 or 100 credits per 10 responses on the basic, standard and premium tiers, so roughly $0.0008, $0.012 and $0.08 a response.
Now cost a conversation rather than an action, because that is the unit your business thinks in. Microsoft's own worked example is a support agent averaging four classic answers and two generative answers per run, which is 8 credits, about $0.064 a conversation. At 900 conversations a day that is 7,200 credits daily. Build the same sum from your own agent's design, and be careful with grounding: Microsoft notes that a tenant-graph-grounded agent can use 12 credits for a single complex prompt, 10 for the grounding plus 2 for the generative answer, which is six times the ungrounded figure.
The reasoning model multiplier is the line most likely to break a forecast built this way. When an agent uses a reasoning-capable model, Copilot Studio bills two meters: the feature rate for the operation, plus text and generative AI tools at the premium rate of 10 credits per 1,000 tokens. So a generative answer that costs 2 credits without reasoning costs 2 credits plus 10 credits per thousand tokens with it. Turn reasoning on per topic, deliberately, after testing whether the cheaper path actually fails, rather than accepting it as a default.
Voice is priced like a different product, because it is
The voice rates are where Copilot Studio stops being competitive with the rest of the market, and the gap is large enough that it should change your architecture rather than just your budget. Classic voice with classic orchestration is 10 Copilot Credits a minute. GenAI voice is 35. Premium GenAI voice is 75. At the pack rate of $0.008 a credit, those are roughly $0.08, $0.28 and $0.60 a minute. Core agent activity, meaning classic answers, generative answers and agent actions, is included in those voice rates, which is a genuine simplification.
Set those against the published rates from specialist voice platforms read on the same day. Vapi charges $0.05 a minute for orchestration with providers billed separately or at nothing with your own keys. ElevenLabs Agents works out at $0.08 a minute across every tier. Retell AI publishes an all-in band of $0.07 to $0.31 a minute including the model and US telephony. Bland AI bundles at $0.14 a minute with no monthly fee. GenAI voice on Copilot Studio at about $0.28 a minute sits above all of them, and premium GenAI voice at about $0.60 sits far above.
That does not automatically make it the wrong choice, because those specialist platforms do not ground an agent in your tenant's Microsoft Graph and the comparison is not like for like. It does mean that if voice is the primary use case rather than an extension of an existing text agent, you should price a dedicated voice platform against this before committing. The question to answer is whether the calls genuinely need the tenant data, because if they do not, you are paying a substantial premium for a capability the call never uses.
The practical mitigation, if you stay, is to be deliberate about which voice tier each agent uses. The difference between classic voice at 10 credits a minute and premium GenAI voice at 75 is more than sevenfold on identical call volume. A booking confirmation or an opening-hours query does not need the premium tier. Reserve it for the conversations where the quality difference is doing commercial work, and audit which tier each published agent is actually running on, because defaults have a way of surviving into production.
The enforcement cliff, and how to not fall off it
Most platforms handle overspend by charging you more. Copilot Studio handles it by switching your agents off, and that is a materially different risk to manage. Microsoft enforces purchased capacity monthly, unused Copilot Credits do not carry over, and enforcement is triggered when a tenant reaches 125 percent of its prepaid capacity. At that point custom agents are disabled. Conversations already in progress finish, and subsequent attempts are rejected until capacity is increased or resets. End users see a message saying there is a billing issue, or that the agent has reached its usage limit.
A customer-facing agent showing a billing message is a worse outcome than an unexpected invoice, and it is worth designing around explicitly. Agent flows behave differently again: when prepaid capacity is fully consumed, new agent flow runs are blocked while the parent agent keeps working for classic answers, generative answers and agent actions. Flow authors see a design-time warning that capacity is exhausted, and enforcement resets monthly when credits renew. Test runs in the flow designer or the agent test chat do not consume capacity for agent flow actions and are not blocked.
There are three published ways out and they suit different situations. You can reallocate existing capacity between environments from the Power Platform admin centre. You can buy more credits and assign them to the affected environment. Or you can link an Azure subscription and enable pay-as-you-go billing, in which case enforcement does not apply because overage bills to Azure instead. That last option converts a hard availability cliff into a soft cost overrun, which for anything customer-facing is almost always the right trade.
The double bind is that unused credits do not carry over, so both over-buying and under-buying are penalised. Buying for your busiest month wastes capacity in the quiet ones; buying for the average guarantees you hit enforcement in the peaks. The workable pattern is to size prepaid capacity near your typical month, link a pay-as-you-go meter to absorb the peaks, and use the per-agent monthly consumption limits Microsoft exposes in the admin centre to stop one misconfigured agent consuming the tenant's pool.
At a glance
Copilot Studio published credit rates, read from Microsoft's Copilot Studio pricing page and Microsoft Learn billing documentation on 20 August 2026
| Feature | Published rate | At the $200 per 25,000 pack rate |
|---|---|---|
| Prepaid Copilot Credit pack | $200.00/pack/month for 25,000 credits | About $0.008 a credit |
| Classic answer | 1 Copilot Credit | About $0.008 |
| Generative answer | 2 Copilot Credits | About $0.016 |
| Agent action | 5 Copilot Credits | About $0.04 |
| Tenant graph grounding for messages | 10 Copilot Credits | About $0.08 |
| Agent flow actions | 13 Copilot Credits per 100 actions | About $0.104 per 100 actions |
| Content processing tools | 8 Copilot Credits per page | About $0.064 a page |
| Text and generative AI tools (basic) | 1 credit per 10 responses | About $0.0008 a response |
| Text and generative AI tools (standard) | 15 credits per 10 responses | About $0.012 a response |
| Text and generative AI tools (premium) | 100 credits per 10 responses, 10 credits per 1,000 tokens | About $0.08 a response |
| Classic voice | 10 Copilot Credits a minute | About $0.08 a minute |
| GenAI voice | 35 Copilot Credits a minute | About $0.28 a minute |
| Premium GenAI voice | 75 Copilot Credits a minute | About $0.60 a minute |
| Microsoft 365 Copilot | $30.00 user/month, paid yearly | Zero-rates employee-facing agent usage, subject to fair use |
| Overage enforcement | 125 percent of prepaid capacity | Custom agents disabled, users see a billing issue message |
What mistakes should you avoid?
- Budgeting in credits without converting to the $0.008 pack rate
- Assuming the free Copilot Studio user licence means a free start, when the tenant needs a prepaid pack first
- Costing a text agent and then adding voice at the same rate, when GenAI voice is 35 credits a minute
- Leaving a reasoning model on by default and paying the premium tool rate on every answer
- Buying capacity for an average month when unused credits do not carry over
- Discovering the 125 percent enforcement threshold when customers see the billing issue message
FAQ
- How much does Copilot Studio cost?
- It is sold as tenant-wide prepaid packs of 25,000 Copilot Credits at $200 a pack per month, about $0.008 a credit, as published on 20 August 2026. Pay-as-you-go through a linked Azure subscription bills only for credits used, and a one-year Copilot Credits prepurchase plan buys a pool of commit units upfront.
- What does one conversation cost?
- It depends on the design. A classic answer is 1 credit, a generative answer 2, an agent action 5, and tenant graph grounding 10. Microsoft's own example of four classic and two generative answers per run comes to 8 credits, about $0.064 at the pack rate. A tenant-graph-grounded complex prompt is 12 credits on its own.
- Why is voice so much more expensive?
- Because it is metered per minute rather than per answer. Classic voice is 10 credits a minute, GenAI voice 35 and premium GenAI voice 75, which is roughly $0.08, $0.28 and $0.60 a minute at the pack rate. Core agent activity is included in those rates, but a specialist voice platform will usually be cheaper per minute.
- What happens if I run out of credits?
- Enforcement triggers when a tenant reaches 125 percent of prepaid capacity. Custom agents are disabled, ongoing conversations finish, and later attempts get a message saying there is a billing issue or that the agent has reached its usage limit. Agent flows are blocked as soon as prepaid capacity is exhausted. Pay-as-you-go environments are not subject to that enforcement.
- Does a Microsoft 365 Copilot licence make this free?
- Only for employee-facing use. Microsoft zero-rates classic answers, generative answers and tenant graph grounding for Microsoft 365 Copilot licensed users interacting with agents in Copilot Chat, Teams or SharePoint, subject to fair use. Customer-facing agents get no benefit from those licences at all, and computer-using agent usage is excluded.
Sources & further reading
Done for you
Want this built for your business?
Tell us the workflow. We scope, build, and ship the agent with guardrails and the numbers to prove it worked. The scoping call is free.
Free scoping call. You own the code.
Was this page helpful?

