GitHub Copilot Plans and AI Credits: What You Actually Pay For

GitHub Copilot’s billing changed substantially on June 1, 2026. The old comparison based on 300 or 1,500 “premium requests” is no longer correct for most subscribers. Current monthly plans use GitHub AI Credits, where cost depends on the model and the number of input, output, and cached tokens consumed.

That change makes plan selection less about counting chat messages and more about workload. Inline completions remain simple to understand; a long agent session that reads many files, reasons deeply, and edits a repository can consume much more of the monthly allowance than a short coding question.

The prices and allowances below were verified against GitHub’s official pages on August 8, 2026. They are time-sensitive and should be checked again before purchase.

Current Individual Plans

GitHub’s current Copilot plans page lists four individual tiers:

The paid plans include unlimited code completions and next-edit suggestions. “Unlimited” applies to those editor features; it does not mean unlimited chat, CLI, code review, or agent usage. GitHub can also apply rate limits and product-specific constraints.

GitHub offers Student access for verified students, while Business and Enterprise plans add organizational licensing, policy management, reporting, and other enterprise controls. Team buyers should compare those organization plans separately instead of assigning individual subscriptions to employees.

What an AI Credit Represents

GitHub’s usage-based billing documentation defines one AI Credit as $0.01 USD. Copilot measures the tokens involved in an interaction and converts their model-specific cost into credits.

Three factors dominate consumption:

• Model choice: more capable or expensive models cost more per token.
• Context size: sending a large repository, long chat history, logs, or many files increases input tokens.
• Agent work: an agent may make multiple model calls, inspect files, run tools, review results, and retry within one task.

This means “one prompt” is not a billing unit. Asking for a one-line PowerShell explanation and delegating a multi-file refactor are both single user actions, but they can consume very different amounts.

The current individual allowances are split into base credits and a flex allotment:

Base credits match the subscription price and do not change. GitHub describes the flex allotment as variable because model economics and efficiency can change. Unused included credits do not roll over; the allowance resets at 00:00 UTC on the first day of each calendar month.

What Consumes Credits—and What Does Not

GitHub currently lists these model-backed features as consuming AI Credits:

• Copilot Chat;
• Copilot CLI;
• agent mode and cloud agent workflows;
• Copilot Spaces;
• Spark;
• third-party coding agents;
• other model-driven features identified in the product and billing documentation.

Code completions and next-edit suggestions do not consume AI Credits and remain unlimited on paid plans. Free includes 2,000 completions per month.

Code review deserves separate attention. GitHub’s June 2026 billing announcement states that Copilot code review consumes GitHub Actions minutes in addition to AI Credits. A repository owner should therefore monitor both billing surfaces rather than assuming the credit meter captures the entire cost.

What Happened to Premium Requests?

Before June 1, 2026, Copilot used premium request units and model multipliers. Pro included 300 premium requests per month and Pro+ included 1,500. That information remains relevant only to a limited legacy group.

GitHub’s legacy billing explanation says existing annual Pro and Pro+ subscribers could remain on request-based billing until their annual plan ends. Those accounts still see premium requests and model multipliers. New monthly subscriptions and migrated accounts use AI Credits.

Therefore, two users with “Copilot Pro” can temporarily see different meters:

• a current monthly user sees AI Credits;
• an eligible legacy annual user may still see premium requests until that annual term ends.

Do not compare the two numbers directly. A premium request is an interaction-based legacy unit; an AI Credit represents token- and model-based cost.

What Happens When the Allowance Is Exhausted

GitHub documents three choices for an individual user who runs out of included credits:

1. wait for the monthly reset;
2. upgrade to a higher plan, paying the difference and receiving the larger allowance minus usage already consumed that month;
3. set a budget for additional usage and continue at the metered rate, subject to GitHub’s additional-usage limits.

A $10 additional budget corresponds to 1,000 AI Credits at the fixed conversion of $0.01 per credit. The actual amount of work those credits buy still varies by model and token usage.

Set an explicit budget before starting unattended agent jobs. A spending cap is especially useful for repositories with generated files, large logs, vendor directories, or broad prompts that cause the agent to inspect more context than necessary.

Choosing a Plan for PowerShell and Server Administration

For PowerShell, infrastructure-as-code, configuration files, and administration scripts, the editor workflow matters as much as the model list.

Free is enough for a real trial

Use Free to test completion quality, basic chat, and whether Copilot fits the chosen editor. It is not the best benchmark for daily agent use because chat and agent access are intentionally limited.

Pro is the sensible starting point for one administrator

Pro provides unlimited paid-plan completions, model selection, Copilot CLI, code review, cloud-agent access, and 1,500 monthly AI Credits. It is the practical first paid tier for someone who mostly writes and explains PowerShell in an IDE and delegates occasional multi-file changes.

Start here unless measured usage shows that the credit allowance interrupts normal work. Buying Pro+ in advance only because it has “more requests” uses an obsolete decision model.

Pro+ fits frequent frontier-model and agent use

Pro+ makes sense when the usage page shows regular pressure on Pro’s allowance, or when the work consistently benefits from premium models and longer agent sessions. Its 7,000-credit total is more than four times Pro’s current total allowance, but the $39 subscription also costs nearly four times as much.

Max is for sustained agent workloads

Max targets heavy, ongoing agent work. A solo administrator who mainly uses inline suggestions and short chat sessions is unlikely to need it. A developer repeatedly delegating large repository tasks may prefer its 20,000 included credits and higher-volume positioning.

Reduce Credit Use Without Reducing Quality

Credit efficiency is mostly context discipline:

• open a new conversation when old history is no longer relevant;
• exclude generated files, binaries, dependency trees, backups, and large logs from agent context;
• provide the exact error window instead of an entire diagnostic archive;
• use a lighter model for syntax, formatting, and routine explanations;
• reserve higher reasoning settings and large context windows for genuinely complex debugging;
• break a vague repository-wide task into bounded, verifiable stages;
• review the usage dashboard after representative work, not after an artificial one-prompt test.

GitHub notes that larger context windows and higher reasoning levels consume more credits. Auto model selection provides a documented 10% model-cost discount on paid plans in supported Copilot surfaces, which can be useful when a specific premium model is not required.

Security Matters More Than the Plan Name

Server administration prompts can contain production hostnames, IP addresses, logs, connection strings, and credentials. Remove secrets before sending context to any coding assistant. Use repositories and test data that follow the organization’s data-handling rules.

GitHub says Business and Enterprise data is not used to train its models. For individual subscriptions, GitHub may use interaction data for model improvement unless the user opts out in Copilot settings. Review the current settings and policy documentation before using an individual plan with company code.

Copilot can propose PowerShell commands, but it does not convert an unsafe production change into a safe one. Require review, backups, limited privileges, and a rollback path for scripts that modify Active Directory, firewalls, storage, virtualization, or operating-system configuration.

The Practical Buying Rule

For one administrator working mostly in VS Code, begin with Free for evaluation and move to Pro for regular use. Watch actual AI Credit consumption for a full representative month. Upgrade to Pro+ only when complex models or agent sessions repeatedly exhaust Pro, and consider Max only for sustained high-volume delegation.

Most importantly, ignore old tables that present premium requests as the current universal billing model. In August 2026, AI Credits are the default, and workload size—not simply the number of prompts—determines how quickly the allowance is used.

Related Guides

• AI Search vs Coding Agent for Linux, PowerShell, PHP, and MariaDB Work
• How to Choose an AI Coding Assistant for WordPress in 2026