GitHub Copilot AI Credits Explained: New Billing, Budgets and Copilot Fusion

Last updated: 9 November 2026

How AI Credits work

GitHub moved Copilot to usage-based billing during 2026. Instead of a fixed number of “premium requests”, most AI features now consume AI Credits based on the models and work involved. One credit equals $0.01.

Plans and included credits

A useful rule of thumb from analysis of these prices: Standalone is cheaper than Business only if average monthly usage stays below about $9, and cheaper than Enterprise only below about $29. Heavy users are usually better off on plans with included credits. For individual plans, see GitHub Copilot pricing.

The GitHub Enterprise bundle cannot be combined with Business or Enterprise seats in the same organisations, is not available for GitHub Enterprise Server, and will open to self-serve customers in stages.

Budgets and cost controls

Budgets can be set in US dollars at the user, cost centre, organisation and enterprise levels. How they behave matters:

GitHub does not publish a default budget amount and warns that agent mode and the cloud agent make several model calls per task. The gh aw forecast command estimates monthly AI credit spend for Agentic Workflows.

Code review costs

Admins can cap the effort level. A team running Max reviews on every pull request could spend far more than expected, so set Balanced as the default and reserve Max for critical changes. Copilot approvals can now count toward required pull request reviews, off by default, with controls over which file paths it can approve.

Copilot Fusion

Copilot Fusion became generally available on 28 October 2026. It is a model picker option that selects models and orchestration for each task, using three workflows: Single, Cascade and Critique. There is no separate Fusion fee; you pay the standard rate of each model it calls.

• Paid plans get 30% off model costs until 30 November 2026. GitHub’s documentation does not say what happens afterwards; without the discount, the same usage would cost about 43% more.
• GitHub claims Fusion Max scores 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code with Opus 5.5 Max, at about half the cost and tokens per task. Artificial Analysis confirmed the cost and token figures ($5.59 versus $11.42 per task) but not the time claim.
• Some users reported Fusion missing from their accounts shortly after launch, so availability may vary during rollout.

For how routing works in general, see model routing explained.

Which models can Copilot use?

Copilot offers several providers’ models. OpenAI’s GPT-6.1 Terra is rolling out across Copilot on paid plans, for example. Model choice affects credit usage, so pick cheaper models for routine work. See GPT-6.1 Terra and the best AI coding assistants.

How to keep Copilot costs predictable

1. Set user-level budgets as hard stops for most developers, with higher limits for power users.
2. Enable “Stop usage when budget limit is reached” on enterprise budgets if you need a firm cap.
3. Cap code review effort at Balanced by default.
4. Set limits on Automations and Agentic Workflows, and forecast with gh aw forecast.
5. Review usage weekly during the first month on new billing.
6. Use cheaper models for simple chat and routine tasks.
7. Plan for 1 December, when the 30% Fusion discount period ends.

Other GitHub Universe 2026 announcements

• Agentic Workflows are generally available, running read-only by default in a sandboxed container.
• Agent Merge lets Copilot fix failing workflows and merge conflicts in the cloud.
• The GitHub Copilot SDK will become the Microsoft Copilot SDK, with no date and no action needed now.

Copilot vs alternatives on cost

Usage-based billing makes comparisons harder, because cost depends on how much agent work you do. Compare your real usage across tools rather than list prices. See Cursor vs GitHub Copilot, Claude Code vs Codex and how to reduce Claude Code costs.

Cost scenarios for a small team

These illustrative scenarios show how credits behave. Real usage varies by model, task size and agent activity.

Setting up budgets: a quick walkthrough

1. Open your enterprise or organisation billing settings on GitHub and find Copilot usage and budgets.
2. Create a default user budget that covers normal use, for example a few dollars above the included credits per person.
3. Create higher budgets for named power users or cost centres.
4. Turn on “Stop usage when budget limit is reached” at the enterprise level if you need a firm ceiling.
5. Cap code review effort in Copilot policy settings.
6. Set limits on Automations per run, day, week or month.
7. Add alerts and assign someone to review usage each week for the first month.

Reducing credit usage without slowing developers

• Lean on completions for routine code, since they are not billed in credits.
• Use smaller models for quick questions and explanations.
• Give agents clear, scoped tasks so they make fewer exploratory calls.
• Keep repositories tidy with good README files and instructions, so agents find context faster.
• Review agent runs that cost the most and see what went wrong.
• Use Fusion’s Single workflow for simple tasks and reserve Critique for important changes.

What changes on 1 December 2026

The 30% discount on model costs for paid plans runs until 30 November 2026. From 1 December, unless GitHub announces otherwise, the same usage costs more. Before then:

• Record your average daily credit use in November.
• Multiply by about 1.43 to estimate December costs without the discount.
• Adjust budgets and model choices if the result is above your plan.

Frequently asked questions