Copilot Credits Default On Dec 1 for CSP Buyers, So Set Caps Now

What Copilot Credits cost against the seat

At $0.01 each, 3,000 credits equals $30. That's our arithmetic, and it's the most useful number here: on Quisitive's $30 seat, a user who burns 3,000 credits in a month has doubled their Copilot cost. None of these write-ups prices the Business seat separately, so run the same sum against what you pay.

A Guide to Cloud's pricing guide says Microsoft sorts Cowork tasks into light (about 100 to 300 credits), medium (about 400 to 700) and heavy (700 or more). That works out to roughly $1 to $3, $4 to $7, and $7 and up per task. A 3,000-credit cap therefore buys about ten light tasks at the top of their band, or about four 700-credit heavy ones.

Treat those bands as a starting point. The same guide says each task's cost depends on the models it runs, the context it pulls in, the tools it calls and its runtime. Model choice is the lever admins control most directly. Quisitive listed Anthropic's Opus 4.8 and Sonnet 4.6 for Cowork at June's general availability, with GPT-5.5 open to Frontier customers and a cheaper Microsoft-built model, Cowork 1, marked as coming soon. Check the current list in your tenant, since those are June listings.

Microsoft's Message Center notice MC1393471 points admins to a Cowork Cost Estimator. For real numbers, cowork.tips suggests enabling Cowork with a small spending limit and starting on light tasks, which gives you your own consumption data.

What default-on means for your bill

"Default-on" sounds like charges start by themselves, and the write-ups say they don't. The Lalit Blogs says the date doesn't mean credits are charged immediately: enterprise tenants still need an admin to create a spending policy, while the billing infrastructure gets enabled in the background. ChangePilot says nothing is billed in a tenant until an admin creates a policy, and Futurum says the services stay off for enterprise customers until then.

There's one gap. The default-on dates are described for Copilot Business purchases, while the Lalit and Futurum assurances name enterprise tenants and customers. ChangePilot's wording is general, but none of these write-ups says outright how a new Business tenant bought through a CSP behaves on December 1. If that's you, get the answer from your CSP in writing.

Why this matters more than the credit price

Copilot used to be a simple procurement sum: seats times the license price. Now part of the cost depends on which group runs which model at what effort level. Futurum's Keith Kirkpatrick frames the trade this way: buyers get a predictable floor for broad deployment, and the hardest forecasting problem, the cost of long-running agents, lands on enterprise FinOps teams.

ChangePilot says spending policies can assign models and effort levels by group, including which models Auto may select in Cowork. That's the setting I'd spend the most time on. Auto weighs accuracy, speed and cost on each request, so whoever defines its allowed models has a direct hand in the budget.

ChangePilot also mentions Viva Insights dashboards that connect credit consumption to completed work. They'll help, but they're Microsoft's own product grading spend on Microsoft's own meter. Pair them with a measure the business already trusts, such as cycle time or tickets closed.

If you also pay for GitHub Copilot, A Guide to Cloud says its GitHub AI Credits can't be spent on Cowork, and that the Microsoft Agent Pre-Purchase Plan is the one prepaid plan covering both. The sticker price rarely tells you the bill with metered AI, a gap we also covered in what AI customer service agents cost small businesses.

What to check before December 1

If you buy through a CSP, you have a month. If you buy direct, these checks are already overdue.

1. Read MC1479276 in your tenant. Confirm whether the default applies to your license type and purchase channel, and get your CSP's answer in writing.
2. Create spending policies by security group. The Lalit Blogs places the controls under Billing, then Cost Management, in the Microsoft 365 admin center, while ChangePilot places them under Copilot, then Cost management. Start with a small pilot group and a low monthly cap.
3. Choose what users without a policy see. ChangePilot says admins now pick between "Request access" and "Learn more", replacing the old all-or-nothing toggle for hiding these features.
4. Restrict models before you widen access. Decide which groups get frontier models and what Auto can choose in Cowork. Staff using Copilot in Teams for meeting recaps have no obvious need for the top model.
5. Route the 75% and 90% alerts to a budget owner. An alert that lands only with IT won't change anyone's behavior.
6. Test what a cap does to long jobs. The fallback described by The Lalit Blogs covers users moving to a standard model. Find out what happens to a multi-step Cowork task that hits the limit partway through, before someone's deliverable depends on it.
7. Keep routine automation off the meter where you can. A repetitive workflow may not need a frontier model at all, and open source Zapier alternatives can be a cheaper home for that work if you can self-host.

If you're also weighing a bundle deal, the Microsoft 365 E7 discount that ends December 31 lands in the same window. Ask Microsoft or your reseller how credits are treated under that agreement, and model them as a separate line until someone shows you otherwise in writing.

Sources

• ChangePilot, Copilot Credits and usage-based billing under MC1479276, 15 October 2026
• The Lalit Blogs, FinOps guide to Microsoft 365 Copilot Credits
• Quisitive, Copilot Cowork pricing and usage-based billing explainer
• Handsontek M365 Admin, Cowork in Frontier and usage-based billing, 17 June 2026
• Futurum, Microsoft Splits the Copilot Pricing Model Into Everyday and Advanced AI, 30 September 2026
• Solv Systems, Microsoft Copilot pricing: licenses and Copilot Credits
• A Guide to Cloud, Microsoft Copilot Cowork pricing guide
• M365 Change Digest, MC1393471
• cowork.tips, Copilot Cowork pricing explained