M365 Copilot Agents Cost Calculator

★ Agent Type

What type of agent are you estimating costs for?

Custom agent
Copilot Studio Agent Builder
M365 Copilot SharePoint agent Foundry Agent
Azure AI

F Foundry Agent Configuration

Foundry agents are billed by tokens (input + output) via Azure. Configure your model and conversation pattern below.

📚 Microsoft Foundry Agent Service (Microsoft Learn)

Model & Token Configuration

Model

System prompt size (tokens)

Avg user message (tokens/turn)

Avg assistant response (tokens/turn)

Turns per conversation

Foundry Built-in Tools

File Search calls per conversation ($2.50 / 1K calls)

Code Interpreter sessions per conversation ($0.033 / session)

ⓘ Web search (Bing Grounding) and MCP server calls incur additional charges not modeled here. See Bing Grounding pricing and the Foundry tool catalog for details.

1 Knowledge Sources

Which knowledge sources does your agent use? Select all that apply.

📚 Supported knowledge sources (Microsoft Learn)

SharePoint / OneDrive

sites:

12 cr/query (with graph)

Dataverse

tables:

2 cr/query (generative answer)

Public websites

URLs:

0 cr (free — not graph-grounded)

Uploaded files

files:

2 cr/query (generative answer)

Enterprise data (Graph/Copilot connectors)

connectors:

12 cr/query (with graph)

ⓘ Source counts (sites, tables, URLs, files, connectors) are for planning reference and are included in the CSV export — they do not affect credit calculations. Billing is determined by query type and turn mix, not by the number of connected sources.

2 Topics, Tools, Prompts & Flows

How many of each does your agent have?

Authored (classic) topics (1 credit when triggered)

Tools / Connectors (5 cr per action + 2 cr gen answer)

AI Prompt actions (text & generative AI tools — see tier below)

Text & generative AI tools tier

Agent flows (7 cr trigger + 13 cr/100 actions)

Actions per flow run (avg. steps in each flow)

Uses reasoning model +10 cr per generative answer and agent action

+ Content Processing (Optional)

Does your agent process documents or images?

Pages processed per conversation (8 cr per page)

3 Typical Test Conversation

How many user turns does a typical test conversation have, and what happens in each turn?

📚 Billing examples (Microsoft Learn)

User turns per conversation (sum of turn types below)

Of those turns, how many involve…

Knowledge lookups (searches a knowledge source)

Tool / Connector calls (calls an action — 5 cr + 2 cr gen answer)

AI Prompt runs (runs a prompt action)

Classic topic responses (authored, no AI)

Agent flow runs (7 cr trigger + 13 cr/100 actions)

4 Test Plan Scale

How many test scenarios and iterations do you plan to run?

Users with a Microsoft 365 Copilot license typically consume zero credits for eligible authenticated usage. Use the `% users with M365 Copilot license` field below to model that blended scenario.

Distinct test scenarios

Iterations per scenario

Monthly prepaid credits (optional)

Pricing model

% users with M365 Copilot license

5 Example Prompt & Cost Trace

Here is what a single test conversation costs, step by step, based on your configuration.