GH-700 study guide: Copilot budgets, AI credits and policies

"Use GitHub platform features" is one of the four skill areas of the GH-700 exam and carries 20–25% of the score. This lesson covers how GitHub Copilot usage is billed, which budget stops what and when, how a manager sees usage, how the model changes cost, and which controls govern Copilot. Questions describe a situation, such as a blocked team or a bill that ran over, and ask which setting caused it or would fix it.

Now you can get Claude Certified and earn OpenAI badges and certifications too → Here's how

What the official study guide covers

How usage-based billing works since June 2026

On 1 June 2026 GitHub replaced request-based billing ("premium requests") with usage-based billing. Every Copilot interaction consumes tokens, the small pieces of text a model reads and writes: input tokens (what is sent, including files and history), output tokens (what the model writes) and cached tokens (text the model already read earlier in the same conversation and does not need to read again in full). Each kind has a price per million tokens that depends on the model, and GitHub converts the total into AI credits, where 1 AI credit = US$0.01. So cost depends on which model ran and how many tokens it used.

Each Copilot Business licence includes 1,900 AI credits a month and each Copilot Enterprise licence 3,900. These are pooled, not personal: 100 Business users share one pool of 190,000 credits, so heavy users draw on what light users leave. Unused credits do not carry over; the pool resets at 00:00 UTC on the first of each month.

When the pool runs out, the AI credits paid usage policy decides what happens. It is enabled by default, so usage continues at US$0.01 per credit, charged to the organization or enterprise. If an administrator disables it, usage is blocked until the pool resets.

Budget guardrails: which one stops what, and when

A month has two phases: the pool phase, while credits remain, and the metered phase, once the pool is used up and paid usage runs.

A cost center is a group of users an enterprise sets up for billing, such as a department. A user-level budget can be universal (one default for everyone), per cost center, or for one named person.

"Stop usage when budget limit is reached" is off by default. Without it, those three budgets only send a notification while charges keep growing.

For each request, GitHub checks in this order:

1. User-level budget. If it is used up, the request is blocked. If a person is covered by more than one user-level budget, the most specific one counts: an individual budget beats a cost center one, which beats the universal one.
2. Shared pool. If credits remain, the request is served from the pool.
3. Metered budget. If the pool is empty, GitHub checks the user's cost center budget, or else the organization budget, or else the enterprise budget.

Two rules follow. The budget with the least money left blocks first: a user with US$5 left is still blocked if the enterprise budget, with "Stop usage" on, has US$1 left. And raising a team or enterprise budget does not unblock someone who has hit their user-level budget.

The enterprise budget caps metered charges, not the whole bill. With 400 Business licences at US$19 (US$7,600) and a US$5,000 enterprise budget that has "Stop usage when budget limit is reached" turned on, the maximum bill is US$12,600.

A blocked user does not get a cheaper model. Features that use credits stop; code completions keep working. The block lasts until the next cycle, or until an administrator raises the budget or approves a request the user submits for more budget (administrators see these under Requests from members in billing settings).

GitHub recommends a universal user-level budget above the licence value (US$19 Business, US$39 Enterprise), so that heavy users can draw on credits that light users leave unused; individual overrides for heavy users, with an expiry date when the override is temporary; an enterprise limit with "Stop usage" on, and a monthly review.

In Copilot CLI, /limits set max-ai-credits 50 caps one session (minimum 30 credits, public preview). It is a soft limit: a response in progress finishes first.

How a manager sees token usage

An enterprise administrator opens Billing & licensing, then AI usage. The chart groups consumption by model and can be filtered by user, organization or cost center, then exported. A second page, Metered usage, shows paid spending over time; type product:copilot in its filter to show Copilot only.

Each user sees their own total in Copilot settings, under Usage this cycle.

These views show credits by user and model, not by prompt; GitHub says you generally cannot trace credits to a specific prompt. In Copilot CLI, /context shows current context use and /chronicle cost-tips suggests ways to cut token use (chronicle is Copilot CLI's record of past sessions).

How the model affects output and cost

At the rates listed on 26 October 2026, a task with 200,000 input tokens, 20,000 output tokens and no caching costs:

A stronger model is not always better. GitHub advises powerful models for design and hard debugging, versatile models for carrying out a clear plan, and lightweight models for routine changes. Switching models mid-session costs more: cached tokens, typically billed at 10% of the input price, cannot pass to another model, so the context is billed again.

Auto model selection lets Copilot choose. A router sends each prompt to the model that can handle it most efficiently, avoiding busy models and changing model only where the saved context would not be wasted. On paid plans it cuts model costs by 10% in Copilot Chat, Copilot CLI, the GitHub Copilot app and cloud agent. In VS Code, Copilot CLI and the Copilot app you can pick a tier: Efficiency (cost first), Balance or Intelligence (quality first). Auto only uses models your plan and policies allow.

For explainability, meaning being able to tell how an answer was produced, you can see which model wrote each response: hover over it in Chat or the Copilot app, read it in the CLI terminal, or find it at the end of a cloud agent response. Copilot code review is the exception: its model is not disclosed.

Governance, privacy and risk controls

Policies control which Copilot features, agents and models people can use. The enterprise sets each policy first (enabled, disabled or "let organizations decide"); an organization cannot override an enabled or disabled setting, while "let organizations decide" hands the choice to the organization. If a user gets licences from several organizations in one enterprise, the least restrictive setting applies, with a few exceptions GitHub lists; across different enterprises, the most restrictive applies, again with a few listed exceptions. Models not covered by GitHub's data retention agreement stay disabled until an administrator enables them.

Content exclusion tells Copilot to ignore chosen files, such as a folder of contract templates. Repository administrators, organization owners and enterprise owners can set it on Business and Enterprise plans. Excluded files do not inform suggestions or answers. It is not supported in Edit and Agent modes of Copilot Chat in VS Code and other editors, and it does not apply to symbolic links (shortcut files that point to another file).

The audit log records changes to Copilot settings, policies and licences, and agent actions on GitHub. Search action:copilot for plan events and actor:Copilot for agent actions. It keeps 180 days, so GitHub recommends streaming it to a SIEM (a security monitoring system). It does not hold prompts users type into Copilot locally.

Rules that decide exam answers

1. Among budgets, only user-level budgets act while the pool has credits; cost center, organization and enterprise budgets act only after it is empty.
2. Cost center, organization and enterprise budgets only alert unless "Stop usage when budget limit is reached" is on.
3. The budget with the least money left blocks first; a larger team budget does not unblock a user who hit their own budget.
4. Maximum bill = licence fees + enterprise budget, if "Stop usage when budget limit is reached" is on.
5. A blocked user gets no cheaper model; code completions still work.
6. Cost = tokens times the model's price. Use the smallest model that does the job.
7. Usage is on the AI usage page; settings changes are in the audit log.
8. An organization cannot override an enterprise policy that is set to enabled or disabled.

Two sample questions

These are original Timo practice questions, not official exam questions.

Question 1

An enterprise sets a US$3,000 enterprise budget for Copilot and leaves every other setting at its default. In the third week of the month the shared pool is used up and metered charges reach US$3,000. What happens next?

A. All Copilot features that use AI credits are blocked until the pool resets.B. Users are moved to a lower-cost model for the rest of the month.C. Administrators are notified, and metered charges continue past US$3,000.D. Each user's remaining included credits are used before any block applies.

Answer: C. "Stop usage when budget limit is reached" is off by default, so the budget only sends a notification and charges keep accruing. A would need that setting on. B fails because there is no automatic fallback to a cheaper model. D fails because included credits form one shared pool, which is already empty.

Question 2

A finance manager wants to see which cost centers used the most AI credits last month and on which models, as a spreadsheet. Where should the enterprise administrator look?

A. The enterprise audit log, searched with action:copilotB. Billing & licensing, then AI usage, filtered and exportedC. Each user's Usage this cycle card on their Copilot settings pageD. The Copilot policies page under the AI controls tab

Answer: B. The AI usage page groups consumption by model, filters by cost center and exports. A records settings changes and agent actions, not consumption. C shows one person's usage. D sets which features and models are allowed and reports no usage.

Build exercise

You need a Copilot plan that includes Copilot CLI and a free test repository.

1. Open Copilot settings from your profile picture and note the credits shown under Usage (Business and Enterprise plans show a Usage this cycle card).
2. Start Copilot CLI in a copy of your test repository on your computer (a clone) and run /limits set max-ai-credits 30.
3. Ask Copilot to write a README. Note which model the terminal shows, then run /context.
4. Ask for a contributing guide and an issue template. If the session reaches 30 credits, Copilot stops and asks you to reset the limit. Run /limits unset, then compare Copilot settings with step 1.
5. On paper, plan budgets for 50 Business users: a universal user-level budget above US$19, an enterprise limit with "Stop usage" on, and the maximum bill (US$950 plus your limit).

Practise this topic

Start with the free GH-700 practice test of 20 questions. The paid GH-700 practice tests have 200 questions in four timed 50-question mocks (US$15), and the GH-700 prep material has one module per skill area with guided labs (US$15), or US$25 for both together: see GH-700 practice tests and prep material. The GH-700 exam guide explains the exam and its four skill areas.

Now you can get Claude Certified and earn OpenAI badges and certifications too

• Register for all four Claude certification exams through Amotion AI, which is Certified in the Claude Partner Network. Claude Certified Associate covers business and knowledge work, not coding.
• Take 11 OpenAI PartnerU learning programs with assessments and course-completion certificates, and work towards the expert badges for ChatGPT, Codex and APIs.
• Practise with 20 full-length Claude mock exams.

Join Timo for Claude and OpenAI certifications

Sources

• Study guide for Exam GH-700 (Microsoft Learn)
• Usage-based billing for organizations and enterprises
• Budgets for usage-based billing
• Getting started with budget controls
• Managing requests for additional Copilot budget
• What changed with Copilot billing
• Managing your company's spending on Copilot
• Monitoring your AI Credits usage
• Setting an AI credit session limit
• Models and pricing for GitHub Copilot
• Optimizing your AI usage
• About Copilot auto model selection
• Supported AI models in GitHub Copilot
• About default availability of Copilot features and models
• Copilot policies for enterprises and organizations
• Content exclusion for GitHub Copilot
• Reviewing audit logs for GitHub Copilot

Independent Timo practice material. Not affiliated with or endorsed by GitHub or Microsoft.