Model your Microsoft Copilot Cowork credit usage
Estimate how your organization might consume Copilot Credits when using Cowork. Enter your expected usage, model different scenarios, and get a directional view of potential monthly costs.
Start at the top and follow each step in order. Adjust the assumptions to reflect your organization’s needs. These estimates assume Anthropic Opus 4.8.
| Role | Number of users |
|---|---|
| Corporate Knowledge Workers | |
| Customer-Facing Knowledge Workers | |
| Technical Workers | |
| Managers & Senior Leaders |
Estimate light, medium, and heavy prompts for each role.
The default numbers are based on Microsoft Frontier customer usage as of 27 May 2026. Please update to suit your organization.
| Role | LightNarrow context, lightweight model, 0–1 tool calls, minimal runtime — 0–1 deliverables. | MediumRicher context, capable model, several tool calls, moderate runtime — 2+ outputs. | HeavyBroad context aggregation, high-quality model, many tool calls, sustained runtime — many outputs. | Monthly Copilot Credit spend per user |
|---|---|---|---|---|
| Corporate Knowledge Workers | 0 | |||
| Customer-Facing Knowledge Workers | 0 | |||
| Technical Workers | 0 | |||
| Managers and Senior Leaders | 0 |
The default numbers are based on Microsoft Frontier customer usage as of 27 May 2026. Please update to suit your organization.
| Copilot Credits used per prompt1 | Credits |
|---|---|
| Light | |
| Medium | |
| Heavy |
Important disclaimers
1 The information presented herein is derived from aggregated and anonymized data collected from customers who participated in the Cowork Frontier program. This data is intended for illustrative and informational purposes only and may not reflect the experience or results of all customers. It should not be relied upon as definitive, complete, or predictive of future outcomes.
2 This tool is designed to help inform budgeting considerations for potential Cowork purchases. The data represents generalized estimates and may vary based on individual customer scenarios. Microsoft does not guarantee the accuracy of this information and disclaims any warranties or assurances related to expected results or outcomes.
This model is provided for illustrative purposes only to demonstrate an approach; organizations should build and validate their own models based on their specific needs. Pricing, credit consumption, and model assumptions are subject to change by Microsoft.
What do real Cowork tasks cost?
Instead of abstract “light / medium / heavy” prompts, here’s a directional view of common Cowork jobs — each positioned against Microsoft’s own light / medium / heavy credit values, based on how much context, tool use and revision the job typically needs. Use them to sanity-check the prompt mix you entered above, then read how these are made below before relying on them.
| Cowork task | What’s involved (context, output, tool calls) | Tier | Est. credits | Est. cost |
|---|---|---|---|---|
| Ask a quick question“Explain this clause / summarise this thread” | ~3–8K tokens in, <1K out, 0–1 tool calls, seconds of runtime. | Light | ~125 | ~$1.25 |
| Summarise a document10–20 page document or report | ~15K tokens in, ~1K out, 1 retrieval tool call. | Light | ~200 | ~$2 |
| Draft an email or messageFrom a thread or short brief | ~5K tokens in, ~0.8K out, 0–1 tool calls. | Light | ~150 | ~$1.50 |
| Meeting recap + actionsFrom a transcript | ~20K tokens in, ~2K out, 1–2 tool calls, structured output. | Medium | ~350 | ~$3.50 |
| Analyse a spreadsheetTrends + a short written summary | ~25K tokens in, ~2K out, 2–3 tool calls (data + calc). | Medium | ~450 | ~$4.50 |
| Author a reusable SkillDefine + test 2–3 times | ~20K tokens in, ~3K out, several tool calls across test runs. | Medium | ~900 | ~$9 |
| Draft a PowerPoint deck10–12 slides, first pass | ~50K tokens in (brief + sources), ~6K out, outline + slide-build + layout tool calls, sustained runtime. | Heavy | ~1,200 | ~$12 |
| PowerPoint + 3 iterationsFirst pass then 3 rounds of edits | 1 heavy first pass (~1,200) + 3 medium refinements (~500 each). | Heavy | ~2,700 | ~$27 |
| Run a Skill → branded assete.g. branded one-pager or image, per run | ~15K tokens in, ~2K out + asset/image generation tool calls (expensive) and runtime. | Heavy | ~1,000 | ~$10 |
| Complete a BRDBusiness Requirements Document | ~60K tokens in (templates + stakeholder docs), ~8K out across sections, many tool calls + 2 iterations. | Heavy | ~2,000 | ~$20 |
| Build a business caseNarrative + simple financial model | ~60K tokens in, ~8K out, calculation + data tool calls, 2–3 iterations. | Heavy | ~2,400 | ~$24 |
| Market / competitive briefWeb research + synthesis | ~40K tokens in, ~5K out, many web/retrieval tool calls. | Heavy | ~1,500 | ~$15 |
Directional planning figures — not Microsoft pricing. They are built from Microsoft’s published light / medium / heavy credit values, not from raw model-token costs. Actual consumption varies with document size, iterations and tool usage — always validate against your Microsoft 365 admin center Cost Management data.
How these estimates are made — and how confident we are
1. The only hard inputs. Three numbers come straight from Microsoft’s own Frontier-program defaults, shown in the calculator above: a light prompt ≈ 125 credits, medium ≈ 500, heavy ≈ 1,200. The estimator also uses ~100 Copilot Credits ≈ US$1 (list) — the same conversion as Microsoft’s spreadsheet. Everything in the table is built from these anchors.
2. Each task is positioned relative to those anchors. We classify a job by the work it asks the AI to do — how much context it must read, how many model turns and tool calls it makes, how much it writes, and how many revision rounds are typical (assuming Anthropic Opus 4.8). A quick question is a single light prompt (~125). A first-pass slide deck is one heavy prompt (~1,200). A deck plus three rounds of edits is one heavy prompt plus three medium ones (1,200 + 3×500 ≈ 2,700). The multi-step figures are simply sums of the anchors.
3. Credits meter the whole service, not raw model tokens. A Copilot Credit pays for Microsoft’s complete managed experience — model inference across many internal turns, tool calls, retrieval, orchestration and Microsoft’s own margin. So you cannot back-compute a task’s credits from the published per-token price of any model (raw model cost is a small fraction of the credit cost). The token and tool-call notes in the table describe the shape of the work — why one job costs more than another — not a billing formula.
4. Where we’re least sure. Single-prompt tasks anchored directly to Microsoft’s values (quick question, summary, first-pass deck) are the most reliable. Multi-step heavy tasks (BRD, business case) depend heavily on document size and revision rounds — treat those as a floor that can climb. Asset and image generation (the “branded asset” row) is the least predictable, because it doesn’t scale with text tokens the way the other rows do.
5. Use it as a sanity-check, not a quote. Run a small real pilot, read your actual consumption in the Microsoft 365 admin center Cost Management dashboard, then update the per-prompt credit values in Step 3 of the calculator to match what you see. These figures are independent estimates from Growthguru.ai and are not endorsed by or sourced from Microsoft.
How Copilot Credits & usage-based billing actually work
Microsoft’s usage-based billing charges you for what you actually use, measured in Copilot Credits. Here are the key things every partner and customer should understand before they buy.
What are Copilot Credits?
Copilot Credits are a common currency for eligible Microsoft AI services billed on usage — today that includes Cowork and the Work IQ API, with more agents added over time.
They complement your fixed subscription licences with a flexible, pay-as-you-go option aligned to real usage. In some cases a licence is the entry point that unlocks an AI service billed by credits.
100 credits ≈ US$1 (list)
What drives credit consumption?
Not every prompt costs the same. Consumption scales with how much work the AI does:
- Light — narrow context, a lightweight model, 0–1 tool calls, minimal runtime.
- Medium — richer context, a capable model, several tool calls, moderate runtime.
- Heavy — broad context aggregation, a high-quality model, many tool calls, sustained runtime and many outputs.
How you’re billed
You’re charged based on actual usage, measured in credits. Admins choose how to fund it from the Microsoft 365 admin center:
- Prepaid credits (P3),
- Pay-as-you-go, or
- existing capacity / Azure subscriptions for billing at scale.
Staying in control of spend
The Cost Management dashboard gives admins centralized control and deep visibility:
- Spending policies — who can consume, how much, and where.
- Limits, alerts and hard caps to prevent overspend.
- Overview tab — real-time usage and remaining capacity.
- Consumption tab — drill into usage by user, group, service or agent.
Official Microsoft resources
Everything here is directional. For definitive guidance, pricing and policy, go straight to the source.
Partner FAQ
Microsoft 365 Copilot Credits — CSP / partner frequently asked questions.
aka.ms/CSPM365CopilotPartnerFAQ ↗Microsoft Learn article
Usage-Based Billing and Cost Management for Copilot Credits — the official overview.
learn.microsoft.com ↗Copilot Credits Guide (PDF)
The full Microsoft Copilot Credits Guide (June 2026) — licensing and credit details.
Download PDF ↗Still have questions?
Copilot Credits, Cowork budgeting and Microsoft incentives can get complex fast. Speak to our AI Gurus — trained on Microsoft programmes, content and incentive guides — to understand exactly what this means for your business.
Speak to our AI Gurus