GitHub Copilot AI Credits: Usage-Based Billing Starts June 1, 2026¶

For / Key Points

For: Individuals and administrators using GitHub Copilot Pro, Pro+, Business, or Enterprise.

Key Points:

On June 1, 2026, Copilot plans move from Premium Request Units to GitHub AI Credits¹.
Base subscription prices stay the same, but agentic sessions and higher annual-plan model multipliers change the real cost profile².
Individuals need to monitor included allowances, while organizations need pooled-credit and budget policies based on the preview bill experience¹.

On June 1, 2026, GitHub Copilot's billing unit changes. Premium Request Units disappear, and usage will be converted into GitHub AI Credits based on input, output, and cached tokens¹.

The practical question is straightforward. What should Copilot users measure, and which budget rules should they define before June 1?

What Changes When PRUs Become AI Credits¶

The core shift is that token consumption becomes the billing basis, not the number of requests. GitHub says the old model no longer fit a product where a quick chat question and a multi-hour autonomous coding session could consume very different amounts of compute while counting similarly under PRUs¹.

GitHub AI Credits calculate input tokens, output tokens, and cached context using the published rate for each model. The conversion rate is fixed: 1 AI credit = 0.01 USD². Code completions and next edit suggestions do not consume AI Credits and remain unlimited for paid plans².

The other important change is the end of automatic fallback when PRUs run out. Under the new model, continued usage depends on remaining credits, additional usage budgets, and administrator policies¹.

From here, users need to watch not only how often they use Copilot, but which model they use and how much context each task processes.

Included Allowances by Plan¶

Base subscription prices are not changing, but included AI Credits differ by plan. For individuals, Copilot Pro includes 1,000 credits per month and Copilot Pro+ includes 3,900 credits, matching their 10 USD and 39 USD monthly prices³.

Plan	Monthly price	Included credits	Promotional allowance	Basic behavior after included credits
Copilot Pro	10 USD	1,000	None	Set an additional usage budget or wait for the next cycle
Copilot Pro+	39 USD	3,900	None	Same
Copilot Business	19 USD / seat	1,900 / user	3,000 / user	Controlled by organization pool and additional usage policy
Copilot Enterprise	39 USD / seat	3,900 / user	7,000 / user	Controlled by organization pool and additional usage policy

For Business and Enterprise, included credits are pooled at the billing entity level⁴. For example, 100 Business seats create a shared pool of 190,000 credits per month. Power users can draw more when needed, while lighter users offset that consumption.

Existing Business and Enterprise customers receive promotional included usage for the first three months of usage-based billing, from June 1 to September 1, 2026⁴. That period should be treated as a measurement window, not just a grace period.

Model Prices Set the Cost Curve¶

Model choice now directly affects spending. GitHub Docs list the following June 1, 2026 rates per 1M tokens².

Provider	Model	Category	Input	Cached input	Output
OpenAI	GPT-4.1	Versatile	2.00 USD	0.50 USD	8.00 USD
OpenAI	GPT-5 mini	Lightweight	0.25 USD	0.025 USD	2.00 USD
OpenAI	GPT-5.4	Versatile	2.50 USD	0.25 USD	15.00 USD
OpenAI	GPT-5.5	Powerful	5.00 USD	0.50 USD	30.00 USD
Anthropic	Claude Sonnet 4.6	Versatile	3.00 USD	0.30 USD	15.00 USD
Anthropic	Claude Opus 4.7	Powerful	5.00 USD	0.50 USD	25.00 USD
Google	Gemini 2.5 Pro	Powerful	1.25 USD	0.125 USD	10.00 USD
xAI	Grok Code Fast 1	Lightweight	0.20 USD	0.02 USD	1.50 USD

Lightweight models such as GPT-5 mini and Grok Code Fast 1 are cost-effective for short edits, explanations, and simple file exploration. Powerful models such as GPT-5.5 and Claude Opus 4.7 can cost more than ten times as much on output tokens.

Anthropic models also include a cache write cost in addition to cached input². For example, Sonnet 4.6 has a 3.75 USD / 1M cache write price, while Opus 4.7 has a 6.25 USD / 1M cache write price.

Why Agentic Sessions Cost More¶

Agentic sessions turn one user request into multiple model calls. When a user asks Copilot to fix a feature, the agent may read several files, call tools, resend prior context, implement changes, and iterate on verification.

That workflow is useful, but billing accumulates across input and output tokens. GitHub Docs explicitly note that agent mode and Copilot cloud agent can consume significantly more usage than a quick chat question³.

The practical optimization levers are narrow.

Separate model tiers: Use Lightweight models for small fixes, explanations, and exploration.
Constrain the search space: Specify target directories, files not to change, and the definition of done.
Stabilize reusable context: Keep repeated rules and assumptions in stable locations to reduce wasteful restatement.

Copilot code review adds a separate wrinkle. It consumes AI Credits for token usage and GitHub Actions minutes when running on GitHub-hosted runners². Self-hosted runners avoid Actions-minute consumption, but the cost moves to the organization's own infrastructure.

What Individuals Should Check in May¶

The first task for individual users is to inspect the preview bill experience when it becomes available in May. GitHub says users and administrators will be able to view projected costs from Billing Overview before the June 1 transition¹.

Annual Copilot Pro and Pro+ users need extra attention. Existing annual plans remain on PRU-based billing until expiration, but model multipliers for annual subscribers increase on June 1, 2026².

Model	Current multiplier	New multiplier
Claude Opus 4.6	3	27
Claude Opus 4.7	7.5	27
Claude Sonnet 4.6	1	9
GPT-5.4	1	6
GPT-5.4 mini	0.33	6
Gemini 3.1 Pro	1	6

When an annual plan expires, the user transitions to Copilot Free with the option to upgrade to a paid monthly plan¹. Users can also convert before expiration and receive prorated credits for the remaining value of the annual plan¹.

Anyone heavily using Powerful models should compare the cost of staying on the annual PRU path with moving to monthly usage-based billing.

Organizations Need Budget Rules First¶

For organizations, the main decision is whether additional usage should continue after the pooled allowance is exhausted. Business and Enterprise accounts can allow continued usage at published rates or block usage until the next billing cycle⁴.

Budget control needs to be designed across multiple layers. GitHub says administrators will be able to set budgets at enterprise, cost center, and user levels¹. At minimum, organizations should define three policies.

Global cap: Set the maximum monthly spend for the enterprise or billing entity.
Department allocation: Track engineering, R&D, platform, and IT usage through cost centers.
User-level guardrails: Prevent runaway usage while defining an emergency exception flow.

The key operational detail is that blocked usage does not automatically fall back to a cheaper model. That makes cost control clearer, but a poorly designed budget can interrupt development work.

Budgets are not only a brake. They become measurement points for identifying where agentic development produces value and where teams are repeatedly spending credits on unfocused exploration.

Pre-Migration Checklist¶

The May preview bill is the last practical planning input before the migration. Individuals and organizations should check the following before June 1.

Review projected monthly costs for heavy users in the preview bill.
Inventory Copilot CLI, cloud agent, and code review usage patterns.
Separate tasks that need Powerful models from tasks that can use Lightweight models.
Define additional usage permissions, caps, and exception workflows for organizations.
Recalculate annual Pro / Pro+ economics after the new multipliers.

This is not only a cost-cutting exercise. It is a decision about which agentic workflows deserve budget because they create enough development leverage.

Summary¶

This change is bigger than a Copilot pricing adjustment. It marks the move from a fixed-price subsidy model toward pricing that more closely follows actual model consumption.

Even if monthly subscription prices stay the same, heavier usage can increase total spending. The opposite is also true: teams that choose models carefully, constrain task scope, and structure reusable context may stay within included allowances.

After June 1, Copilot cost management is no longer only a finance task. Engineering teams need to decide which model belongs to which task, because that choice will shape both agentic productivity and budget control.