GitHub Copilot now charges for each chat, CLI session, and agent based on the tokens it consumes, in GitHub AI Credits worth one cent each. In Business and Enterprise plans, the setting that allows continued spending after exhausting included credits comes enabled by default. Here’s how to cap spending at three levels: per agent session, per user, and across the entire company.
What is the price of GitHub Copilot with AI Credits?
The price of each plan hasn’t changed, but now includes a monthly amount of credits equivalent to what you pay. Starting June 1, 2026, all GitHub Copilot plans bill in GitHub AI Credits instead of premium requests. The cost of each interaction depends on two things: the model you use and the tokens you consume (input, output, and cached).
| Plan | Price | Base Credits | Flex Allotment | Total Monthly |
|---|---|---|---|---|
| Copilot Pro | USD 10 | 1,000 | 500 | 1,500 |
| Copilot Pro+ | USD 39 | 3,900 | 3,100 | 7,000 |
| Copilot Max | USD 100 | 10,000 | 10,000 | 20,000 |
| Copilot Business | USD 19 per user | 1,900 | — | 1,900 per user |
| Copilot Enterprise | USD 39 per user | 3,900 | — | 3,900 per user |
In individual plans, base credits equal your subscription price and don’t change. The flex allotment is an extra that applies automatically when you run out of base credits, and GitHub clarifies it’s variable: they may adjust it as model prices evolve. The figures in the table are what the documentation showed at the time of writing.
In organization plans, included credits are pooled together. A company with 100 Business licenses shares 190,000 credits, so heavy Copilot users benefit from what light users don’t consume. The pool resets at 00:00 UTC on the first day of each month, and leftover credits don’t accumulate.
A calendar detail matters here. Organizations already using Copilot before June received a larger promotional allotment between June and August. That promotion ended on September 1, 2026, so October is the first full month many teams pay with the standard allotment. If you search for the official English table as “GitHub Copilot pricing,” these are the figures the documentation lists today.
Is GitHub Copilot free? What consumes AI Credits and what doesn’t?
In paid plans, only part of Copilot is unlimited: code completions and next edit suggestions don’t consume credits. They keep working even when a user is blocked by budget.
Everything else does consume AI Credits: Copilot Chat, Copilot CLI, the Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents. Copilot code review consumes both credits and GitHub Actions minutes.
What about the free plan? Copilot Free exists, but it’s limited: it includes 2,000 code completions per month and a small AI Credits allotment, with access to models only through auto model selection. To work with agents seriously, you need a paid plan.
Why do agents cost more than chat?
Because the cost grows with the work done. According to GitHub’s documentation, a quick chat question with a lightweight model can cost less than one credit, while a long agent session with a frontier model touching many files costs significantly more. GitHub’s own budget tutorial warns that a single intensive user or an automated agent session can consume a disproportionate share of the shared pool early in the month and leave less for everyone else.
That’s exactly the scenario you want to contain when an agent runs unsupervised overnight.
Model choice is the other lever. In paid plans, GitHub offers a 10 percent discount on model costs when you use automatic model selection in Copilot Chat, Copilot CLI, the GitHub Copilot app, or the cloud agent.
How do you limit AI credits per Copilot CLI session?
Copilot CLI lets you set a credit cap per session. In an interactive session, use:
/limits set max-ai-credits 150
To remove the limit:
/limits unset
If you run the CLI non-interactively, for example in a script or in CI, pass the limit as a flag:
copilot -p "YOUR PROMPT" --max-ai-credits 150
A few details to keep in mind:
- The minimum value is 30 credits.
- These are soft limits: if a response is in progress when the cap is reached, that response completes, so actual consumption can go slightly over.
- In interactive mode, Copilot stops and asks if you want to raise the limit and continue from where it left off.
- In non-interactive mode, execution simply stops when the cap is reached.
At the time of publishing, CLI session limits are in public preview, so GitHub may change how they work. If you also want to restrict what the agent can touch on your machine, check how to enable GitHub Copilot’s local sandbox.
How do you prevent your team or company from overspending?
In Business and Enterprise, there are three default values worth reviewing right now. All three are documented by GitHub.
- Paid usage comes enabled. When the shared pool runs out, consumption continues and charges USD 0.01 per credit. If you want a hard stop at included credits, an admin must disable the “AI credit paid usage” policy in the organization’s or enterprise’s AI Controls settings.
- Budgets don’t stop anything unless you tell them to. Enterprise, organization, and cost center budgets only send a notification by default. To work as a real limit, enable “Stop usage when budget limit is reached” on each one. GitHub’s own documentation says it: without that option, budgets are alerts, not barriers.
- Per-user budget is the most important control. A universal user-level budget limits each person’s total consumption from both the pool and paid usage, and always cuts hard. GitHub recommends setting it above the license value (USD 19 in Business, USD 39 in Enterprise) so the shared pool keeps working. Then you can give individual exceptions to people who use agents frequently, and even set expiration dates: they work for a sprint or a specific incident.
If multiple budgets apply to a user, the most specific wins: the individual overrides the cost center, which overrides the universal. If several limits are active at once, the one with the least available margin blocks first.
How much budget should you assign to GitHub Copilot?
This is where most teams get the number wrong. The enterprise budget is not your total monthly budget: it only limits paid usage once the shared pool runs out.
The example from GitHub’s own documentation: 400 Business licenses cost USD 7,600 per month. With an enterprise budget of USD 5,000, the maximum bill is USD 12,600, not USD 5,000.
For a small team the math is the same. Ten Business licenses give a pool of 19,000 credits, equivalent to USD 190. Anything beyond that gets blocked or charged, depending on how you set the three adjustments from the previous section.
GitHub suggests reviewing the AI usage panel at least once a month and reading it like this:
- Users blocked early in the month: the per-user budget is too tight.
- Paid usage appearing unexpectedly: the pool runs out before month’s end.
- The pool lasts all month and nobody gets blocked: that’s the ideal state.
What happens when someone hits the limit?
They lose access to Copilot features that consume credits until one of three things happens: the next billing cycle begins, an admin raises the budget, or an admin approves a request for additional budget.
There’s no automatic switch to a cheaper model. Completions and next edit suggestions keep working all the time.
Is this the same as Microsoft’s Copilot Credits?
No. Microsoft calls “Copilot Credits” a different system that covers Copilot Studio, Cowork, and Dynamics agents, managed from the Microsoft 365 admin center. GitHub’s community team has clarified they only support GitHub Copilot billing and have no visibility into Microsoft 365 Copilot plans. If you work in GitHub, your unit is the GitHub AI Credit.
What about you? Have you already set up a per-user budget on your team, or is your Copilot still running with default settings?
- Per-user budgets
- I disabled paid usage
- Per-session credit limits in the CLI
- I prefer another option (tell us which)
Comment below or, if this article reached you by email, reply directly to the email: your response gets published here.
Related: What’s the best AI for coding in 2026? We compare Claude Code, Cursor, Codex, and Copilot