Open
AgentCosTest.htmlin any modern browser (Edge, Chrome, Firefox). No account, server, or login required.
This tool produces planning estimates only. It is not a billing commitment, a contractual obligation, or a guarantee of any kind.
The numbers you see are based on simplified assumptions about how an agent behaves. Actual charges on your Microsoft or Azure invoice can and will differ — sometimes significantly — because:
- Runtime behavior varies. The agent may take more or fewer turns, trigger different features, or follow different orchestration paths than you modeled.
- Tenant configuration matters. Licensing, capacity pools, and feature availability affect how billing is applied.
- Model usage is dynamic. Token consumption and credit usage change based on the content of real user conversations.
- Microsoft pricing and licensing terms can change. Rates shown reflect Microsoft Learn documentation as of the date noted at the bottom of the calculator. Microsoft reserves the right to update pricing at any time.
Do not use this tool to make contractual cost commitments to customers, management, or finance. Use it to build intuition, compare architectural options, and establish a planning baseline. Validate against actual Azure and Copilot Studio usage reports once the agent is running in production.
This calculator helps you estimate the cost of running a Microsoft 365 Copilot agent before you deploy it. It models credits and dollars, not vague ranges — so you can plan budgets, compare options, and have an informed conversation with stakeholders.
Think of it like a flight-price estimator: the final fare depends on fuel surcharges, seat selection, and what actually happens at runtime. This tool gives you a solid planning baseline — not a receipt.
Key rule: Outputs are planning estimates only. They carry no contractual weight and should not be treated as a billing guarantee.
- Open
AgentCosTest.htmlin a browser. - At the top, pick a Quick start template from the drop-down (for example, Enterprise FAQ agent).
- Scroll to the bottom — you will see an Example Prompt & Cost Trace and a Production Cost Estimate panel filled in automatically.
- Adjust any number that doesn't match your real situation. The estimate refreshes instantly.
- Click Export CSV to save the results, or Print for a PDF snapshot.
That's it. The template does most of the thinking for you.
Remember: the number you see is a starting-point estimate. Your actual invoice will depend on real usage patterns, tenant configuration, and Microsoft's current pricing at the time of billing.
The tool is a single scrollable page divided into numbered steps. Each step feeds into the final estimate at the bottom.
[Agent Type] → [Knowledge Sources] → [Components] → [Conversation Profile] → [Test Scale]
↓
[Example Prompt & Cost Trace]
[Production Cost Estimate]
At the top right you can toggle between Dark and Light themes.
The ⟳ Start Over button resets everything to defaults.
What to do: Select the type of agent you are building.
| Option | What it is | Billing model |
|---|---|---|
| Custom agent (Copilot Studio) | Full-featured agent with topics, tools, flows, and AI prompts | Copilot Credits |
| Agent Builder (M365 Copilot) | Declarative agent built in Agent Builder — instructions + knowledge only | Copilot Credits (public website grounding is free) |
| SharePoint agent | Auto-created from a SharePoint site | Copilot Credits |
| Foundry Agent (Azure AI) | Azure-hosted agent billed by tokens, not credits | USD tokens via Azure subscription |
Not sure which to pick? If you are building in Copilot Studio and your agent has custom topics or can take actions (send emails, create tickets), pick Custom agent. If you simply connected a SharePoint site and let M365 Copilot surface it, pick SharePoint agent.
Quick start template: Below the agent type buttons there is a drop-down with ready-made configurations. Pick one and the rest of the form fills in automatically. Templates available:
| Template | Best for |
|---|---|
| Enterprise FAQ agent | Internal FAQ scenarios using SharePoint + enterprise connectors |
| HR policy agent | Pure internal knowledge (SharePoint, uploaded files) |
| IT helpdesk with tools + flows | Complex agents with tool calls and automated flows |
| Employee Self-Service — HR agent starter | ServiceNow HRSD + Workday HR scenarios; shared orchestrator flow; mixed M365 Copilot user share |
| Employee Self-Service — IT agent starter | ServiceNow ITSM + Microsoft Self-Help; reasoning model for diagnostic classification |
| General purpose assistant (GPT-4o) | Foundry agents answering questions from uploaded files |
| Code assistant (GPT-4.1) | Foundry agents doing code review / generation |
What are knowledge sources? They are the data stores your agent searches to answer questions. Examples: a SharePoint site, uploaded policy documents, or a public website.
What to do: Check every data source your agent connects to, and set the count (number of sites, files, URLs, etc.) for reference.
Important: The count of sources is for planning reference only — it does not change the credit cost. What matters is the type of source and how many lookup turns you configure in Step 3.
| Source | Cost per query turn | Notes |
|---|---|---|
| SharePoint / OneDrive | 12 credits (with Tenant graph grounding) or 2 credits (without) | Tenant graph grounding enabled by default — leave it on for realistic estimates |
| Dataverse | 2 credits | |
| Public websites | 2 credits (or free for Agent Builder agents) | |
| Uploaded files | 2 credits | |
| Enterprise data (Graph/Copilot connectors) | 12 credits (graph-grounded) | Model these the same as SharePoint with graph on |
Tenant graph grounding: When you check SharePoint or Enterprise connectors, a sub-option appears: "Tenant graph grounding enabled." Leave this checked unless you specifically disabled it in your agent configuration. It adds 10 credits per query but dramatically improves answer quality.
What are tokens? Language models charge by tokens — roughly 1 token ≈ 4 characters. Unlike Copilot Studio, Foundry agents have no concept of "credits"; they charge input and output tokens directly to your Azure subscription.
What to do: Fill in these fields.
| Field | What to enter | Example |
|---|---|---|
| Model | The AI model your agent uses | GPT-4o, GPT-4.1-nano |
| System prompt size (tokens) | How long your agent's instructions are | 500 for a simple agent, 2 000 for a detailed one |
| Avg user message (tokens/turn) | Typical length of a user question | 150 tokens ≈ ~600 characters |
| Avg assistant response (tokens/turn) | Typical length of the agent's reply | 400 tokens ≈ ~1 600 characters |
| Turns per conversation | How many back-and-forth exchanges happen | 5 turns |
| File Search calls per conversation | If your agent searches uploaded files | $2.50 per 1 000 calls |
| Code Interpreter sessions per conversation | If your agent writes and runs code | $0.033 per session |
Context accumulation explained (non-technical): Each time a user sends a new message, the agent re-reads the entire conversation history. So a 5-turn conversation is not 5× the cost of 1 turn — it costs progressively more because each turn carries the weight of all previous turns.
These are the capabilities your agent has, not what it does in a single conversation. Think of this as configuring the agent's feature set.
| Field | What it means | Cost when triggered |
|---|---|---|
| Authored (classic) topics | Pre-written scripted responses (no AI involved) | 1 credit per trigger |
| Tools / Connectors | External API or service calls (create ticket, send email) | 5 credits (action) + 2 credits (AI response) = 7 credits |
| AI Prompt actions | Prompt the AI to summarize, translate, classify, etc. | Basic: 0.1 cr / Standard: 1.5 cr / Premium: 10 cr |
| Text & generative AI tools tier | Quality level of the AI model used in prompts | Basic, Standard, or Premium |
| Agent flows | Automated multi-step task sequences | 7 credits base + 13 credits per 100 steps |
| Actions per flow run | Average number of steps in one flow execution | Used to calculate the per-run flow cost |
| Uses reasoning model | Whether the agent uses an advanced reasoning model | +10 credits on every generative answer and agent action |
Content Processing (optional step below components): If your agent reads or processes documents or images, enter how many pages it handles per conversation. Each page costs 8 credits.
This is where you describe what actually happens during a conversation — not what the agent can do, but what it typically does in a single test session.
User turns per conversation is calculated automatically as the sum of all turn types below it.
| Turn type | What it means | Credits used |
|---|---|---|
| Knowledge lookups | Agent searches a knowledge source | 2 cr (+ 10 if graph-grounded) |
| Tool / Connector calls | Agent calls an external system | 7 cr (5 action + 2 gen answer) |
| AI Prompt runs | Agent runs a prompt action | Depends on tier (Basic/Standard/Premium) |
| Classic topic responses | Agent returns a scripted answer | 1 cr |
| Agent flow runs | Agent triggers an automated workflow | 7+ cr depending on steps |
Example: A conversation with 2 knowledge lookups (SharePoint with graph) + 1 classic response =
2 × 12 cr + 1 × 1 cr = 25 credits per conversation.
Warning banner: If you enter knowledge lookup turns but haven't checked any knowledge source in Step 1, the tool will show a warning. Those turns won't cost anything unless a source is enabled.
What to do: Describe the size of your test plan.
| Field | What to enter |
|---|---|
| Distinct test scenarios | The number of unique conversation scripts or test cases |
| Iterations per scenario | How many times you run each script (for consistency testing) |
| Monthly prepaid credits | Your tenant's monthly credit allowance (0 = not using prepaid) |
| Pricing model | Pay-as-you-go, Copilot Credit pack, Copilot Credit P3, or Microsoft Agent P3 (see below) |
| % users with M365 Copilot license | If some users have M365 Copilot licenses, their interactions cost zero credits |
Total test conversations = Scenarios × Iterations.
Pricing model options (all bill in Copilot Credits — only the dollar conversion changes):
| Option | Effective $/credit | Best for |
|---|---|---|
| Pay-as-you-go (PAYG) meter | $0.0100 | Postpaid via Azure, no commitment |
| Copilot Credit pack (subscription) | $0.0080 | Predictable monthly use; $200/25,000 credits |
| Copilot Credit Pre-Purchase Plan (P3) | $0.0095 → $0.0080 | 1-year commit, 9 tiers (5–20% off) |
| Microsoft Agent Pre-Purchase Plan (P3) (NEW May 2026) | $0.0095 / $0.0090 / $0.0085 | 1-year commit covering both Copilot Studio AND Microsoft Foundry usage; 3 tiers (5/10/15% off) |
Source: Microsoft Copilot Studio Licensing Guide — May 2026 (PDF).
M365 Copilot license impact: If 50% of your users have M365 Copilot licenses, half of all interactions are free. The "% users with M365 Copilot license" field accounts for this in the estimate.
Capacity overage: If prepaid credits are configured, the tool shows a capacity bar. At 125% of prepaid, custom agents are disabled by the platform.
Voice agents (introduced in the May 2026 Licensing Guide) are billed by total call length to the nearest second plus the configured voice orchestration. Voice agents are not modeled in this tool — estimate them separately.
This section is read-only — it is generated automatically from your configuration.
It shows a realistic sample conversation with:
- A user prompt for each turn type you configured.
- A breakdown of what the agent does in that turn.
- The exact credit cost of each step.
- A total for the full example conversation.
Use this to sanity-check your setup. If the trace looks wrong (e.g., the cost per conversation is unexpectedly high or low), review Step 3 and adjust the turn mix.
At the bottom of the page, the results panel shows:
| Section | What it tells you |
|---|---|
| Summary boxes | Total test conversations, credits per conversation, total credits, and estimated cost in dollars |
| Credit breakdown | Per-turn costs itemized by feature type (knowledge, actions, prompts, flows, etc.) |
| Credit distribution chart | Horizontal bars showing which features drive the most cost |
| Capacity impact | How testing compares to your monthly prepaid allocation (if configured) |
| Export CSV / Print | Save or share the results |
| Start Over | Reset everything |
When sharing results: Always communicate clearly that the figures are planning estimates and not contractual cost commitments. Actual production costs depend on real conversation patterns, user volume, tenant configuration, and Microsoft's current pricing terms — none of which this tool can predict with certainty.
Pick a template that is closest to your use case, check whether the turn counts in Step 3 look right, update the test scale in Step 4, and read the dollar total. Done.
That is a Tool / Connector call turn. In Step 3, set "Tool / Connector calls" to the number of times the agent calls external systems per conversation. Each call costs 7 credits.
Use the "% users with M365 Copilot license" field in Step 4. Users with those licenses consume zero credits for all features.
Try all four. Change the "Pricing model" in Step 4 between Pay-as-you-go, Copilot Credit pack, Copilot Credit P3 (9 tiers), and Microsoft Agent P3 (3 tiers, NEW May 2026 — covers Copilot Studio + Foundry). The credit count stays the same; only the dollar total changes.
Check SharePoint / OneDrive in Step 1 and set the number of sites for reference. The credit cost per lookup is still 12 credits (with graph) regardless of how many sites — the per-query cost comes from Step 3's knowledge lookup turns, not Step 1's source count.
Select Foundry Agent as the agent type. The rest of the form changes to the Foundry configuration. Fill in Step F (model, token sizes, turns, built-in tools) and Step 4 (scale). The result will be in USD tokens, not credits.
Export CSV: Downloads a flat file with all inputs and outputs — useful for sharing with finance or project stakeholders, or for tracking estimates over time.
Print: Renders a clean printer-friendly version of the full estimate. Save as PDF for asynchronous sharing.
| Term | Plain-language definition |
|---|---|
| Copilot Credit | The billing unit for Copilot Studio agents. 1 credit = $0.01 (PAYG) or $0.008 (prepaid). |
| Generative answer | An AI-generated response to a user query. Costs 2 credits. |
| Classic answer | A scripted, authored response (no AI). Costs 1 credit. |
| Agent action | A tool or connector call (e.g., create ticket, send email). Costs 5 credits. |
| Tenant graph grounding | Using Microsoft Graph semantic search for knowledge retrieval. Adds 10 credits per query. |
| Agent flow | An automated workflow defined in Copilot Studio. Costs 7 credits per run + steps. |
| Reasoning model | An advanced AI model that thinks through complex problems. Adds 10 credits per response. |
| Token | The unit of billing for Foundry agents. 1 token ≈ 4 characters in English. |
| PAYG | Pay-as-you-go pricing. You pay per credit consumed with no upfront commitment. |
| Prepaid pack | Purchase 25 000 credits for $200 upfront — 20% cheaper than PAYG. |
| Overage enforcement | When a tenant reaches 125% of prepaid capacity, custom agents are automatically disabled. |
- Copilot Credits billing rates (Microsoft Learn)
- Reasoning model billing (Microsoft Learn)
- Cost considerations for extensibility (Microsoft Learn)
- Foundry Agent Service overview (Microsoft Learn)
- Azure OpenAI pricing (azure.microsoft.com)
- Production usage estimator (Copilot Studio)
This tool is provided for planning and estimation purposes only. It does not constitute a contract, quote, invoice, or billing commitment of any kind. Microsoft's actual pricing, licensing terms, and billing behavior are governed solely by the applicable Microsoft Customer Agreement, Product Terms, and the Azure pricing pages in effect at the time of use.
Pricing rates used in this calculator were sourced from Microsoft Learn documentation and Azure pricing pages, last verified in March 2026. Rates are subject to change without notice. Always verify current rates at:
Last updated: March 2026 — aligned with calculator v1.3.0