Interactive tool

What does an agent run cost?

A best-endeavours estimate for document-processing agents built on the GitHub Copilot Harness, from Microsoft's published Copilot Credit billing rates. Use it to set a cost ceiling while consumption reporting rolls out to your tenant.

Your agent, per run

Total words the agent reads and reasons over

Across all documents in one run. Digital text bills at token rates only - roughly 1,000 tokens per 750 words

Reasoning model

Standard - 1.5 credits / 1,000 tokens Premium reasoning - 10 credits / 1,000 tokens

Deep-reasoning agents bill the premium token rate on top of feature rates - the line item most estimates miss

Agent actions per run

Triggers, steps, transitions, the output

Agent flow actions per run

If the agent calls agent flows

Contingency multiplier

The harness bills via models, runtime, context and tools - treat the raw figure as a floor

Runs per month

For the monthly figure and pack sizing

Estimate

A$50 per run

~4,200 credits with contingency

Reading and reasoning (per 1,000 tokens)	2,000
Agent actions (5 credits each)	75
Agent flows (13 credits per 100 actions)	0
Raw estimate	2,075
With contingency	4,150 credits
Cost per run (US$ pay-as-you-go)	US$41.50
Monthly (20 runs)	A$996
25,000-credit packs per month	3.3

For reference: Microsoft's own planning ranges put heavy harness scenarios at 500+ credits per run. If your agent does deep multi-step work across many sources, sanity-check a lower estimate against that floor before relying on it.

Read the article

Estimates only, not Microsoft pricing advice. Rates from the Copilot Credits billing rates table, the AI tools licensing page and the Microsoft Copilot Credits Guide (August 2026); all rates subject to change. A$ figures use the Australian prepaid pack rate of about A$300 for 25,000 credits (A$0.012 per credit); US$ figures use pay-as-you-go at US$0.01 per credit. Word-to-token conversion approximated at 1,000 tokens per 750 words. Digital text bills at token rates only; scanned or image documents processed by a document processing or text recognition model add 8 credits per page, which this estimate does not include. Agents on the GitHub Copilot Harness bill through model, runtime, context and tool factors rather than these meters line by line, so treat the result as a planning floor - that's what the contingency multiplier is for. Microsoft also publishes an official agent usage estimator, but it covers the standard builder rather than the GitHub Copilot Harness.

Take a number into the room, not a guess

The Accelerator runs agent experiments with a cost ceiling agreed before the build starts. If the credits question is stalling a pilot, we can help you set the ceiling.

See the Accelerator Talk to us

Copilot Credit run estimate

FutureAbility - futureability.com/copilot-credit-estimator -

Input	Value
Total words the agent reads and reasons over
Reasoning model
Agent actions per run
Agent flow actions per run
Contingency multiplier

Estimate
Raw credits per run
Credits per run with contingency
Cost per run (AU prepaid pack rate)
Cost per run (US$ pay-as-you-go)
Monthly cost ( runs)

Best-endeavours estimate from Microsoft's published Copilot Credits billing rates (August 2026). Text and generative AI tools 1.5 (standard) or 10 (premium) credits per 1,000 tokens; agent actions 5 credits; agent flow actions 13 credits per 100. Word-to-token conversion approximated at 1,000 tokens per 750 words. A$ at the Australian prepaid pack rate of about A$300 for 25,000 credits; US$ at pay-as-you-go US$0.01 per credit. Digital text bills at token rates only; scanned or image documents add 8 credits per page for document processing, not included here. Agents on the GitHub Copilot Harness bill through model, runtime, context and tool cost factors, so treat this as a planning floor. Estimates only, not Microsoft pricing advice; rates subject to change. Sources: learn.microsoft.com (Copilot Credits billing rates; AI tools licensing); Microsoft Copilot Credits Guide, August 2026.