GitHub Copilot's Freemium Era Is Over

A week ago I wrote about how the all-you-can-eat AI buffet is seemingly coming to an end, and how practically every major AI vendor has been tightening the screws. I noted that Microsoft looked like a relative outlier — still following the all-inclusive (for M365 Copilot) or generous grants (GitHub Copilot's "Premium Requests") models, while competitors were moving to token metering. Well, that didn't last long.

On April 27, 2026, GitHub announced that most Copilot plans will transition to usage-based billing starting June 1, 2026. Premium Request Units (PRUs) are out. GitHub AI Credits are in.

In short: The buffet is now closed. And here's what you actually need to know as a practitioner.

Called it (even if the change is more dramatic than I expected).

Prices stay flat. Value does not.

Let's start with the calm before the chaos: base plan prices are not changing.

• Copilot Pro stays at $10/month
• Copilot Pro+ stays at $39/month
• Copilot Business stays at $19/user/month
• Copilot Enterprise stays at $39/user/month

What changes is what that money buys you, and how fast you can incinerate it with heavy models and agentic workflows.

Under PRUs, "one request" was a fuzzy abstraction and GitHub absorbed a lot of real inference cost behind the curtain.

The revenue Microsoft/GitHub makes from the $10-39 monthly plan might be completely wiped out by the inference required to execute one agentic session, that in turn consumed a single Premium Request!

Copilot Pro at $10/month = 1,000 credits.

Copilot Pro+ at $39/month = 3,900 credits.

For Business and Enterprise, credits are pooled across the organisation. 100 Business seats = 190,000 shared credits/month.

What costs credits (and what still doesn't)

Before panic mode kicks in: code completions and Next Edit Suggestions are not billed in AI credits. They remain included and unlimited on paid plans.

If all you do is let GitHub Copilot autocomplete code in your IDE, June 1 doesn't affect your daily workflow at all. Although, let's be real there - does someone still just do that? 😅

What does consume credits:

• Copilot Chat
• Copilot CLI
• Copilot cloud agent (the coding agent / SWE Agent)
• Copilot Spaces
• Spark
• Third-party coding agents (meaning Codex, Claude, etc when used on GitHub)

Additionally, if you're running your agents on GitHub-hosted runners, GitHub Actions minutes will also be consumed.

And the more agentic your workflow, the more this matters. A quick one-liner question in chat? Fractions of a credit. A multi-hour autonomous agent session chewing through a large codebase? That can be ten thousand credits - or a hundred dollars - in a single session!

Model choice is now a budget decision

Under the new model, every interaction costs tokens, and tokens cost money based on which model you're talking to. The rates are public. Here's a rough summary of where things land:

Lightweight / cheap:

• GPT-5 mini: included
• GPT-5.4 nano: $0.20 input / $1.25 output
• Grok Code Fast 1: $0.20 input / $1.50 output
• Gemini 3 Flash: $0.50 input / $3.00 output

Versatile / middle ground

• GPT-4.1: included
• Claude Haiku 4.5: $1.00 input / $5.00 output
• Claude Sonnet 4.6: $3.00 input / $15.00 output

Powerful / expensive:

• Claude Opus 4.7: $5.00 input / $25.00 output
• Gemini 2.5 Pro: $1.25 input / $10.00 output
• GPT-5.4: $2.50 input / $15.00 output

Note: GPT-4.1 and GPT-5 mini are included models, just like they currently consume Premium Requests at 0x multiplier. Looks like this will not change. This is a nice escape hatch for lighter tasks - or tighter budgets!

The practical takeaway: Use a lightweight model for simple tasks, and save the frontier models for things that actually need them.

This is a habit worth building before June 1, not after.

One of the problems I can see with Copilot right now is that the "model router" ("Auto" option in the UI) is not very smart. It often defaults to a more expensive model when a cheaper one would do just fine, and even with the 10% discount when using it, it racks up quite the bill from June onwards if not addressed. So if you're using "Auto", start paying attention to which models it's picking and whether that's actually necessary for the task at hand.

And if before you got 300 Premium Requests for your $19 USD Business plan, meaning each request was effectively worth $0.063, now you get 1,900 AI credits, which means each credit is worth $0.01.

But whereas you could run an agent for half an hour, producing a small PoC application for you with just one PRU, now you'll easily burn though - at the very least - hundreds of your credits in that same session.

The annual-plan gotcha

Annual Pro/Pro+ subscribers are not moved to monthly usage billing mid-plan, but from June 1, 2026 the model multipliers for annual plans jump hard.

Some examples of how multipliers are changing (current → new):

You read that right! Claude Opus 4.7 goes from a 7.5x multiplier to 27x. If you're on an annual plan and have been regularly reaching for Opus or Sonnet 4.6, your effective "included usage" just got cut by almost 90%!

At expiration, annual subscribers move to Copilot Free unless they choose a monthly paid plan.

Copilot code review is about to start costing you Actions minutes

Easy-to-miss detail: from June 1, 2026, Copilot code review consumes GitHub Actions minutes on GitHub-hosted runners, plus AI credits.

Self-hosted runners don't consume Actions minutes, so if you have heavy code review usage and plenty of Actions to run, that might be worth investigating to save your minutes.

Business and Enterprise get a 3-month cushion

Existing Copilot Business and Enterprise customers get a promotional higher credit allowance for June, July, and August 2026:

• Business: $30 in monthly AI credits (vs. the standard $19)
• Enterprise: $70 in monthly AI credits (vs. the standard $39)

After September 1, it returns to standard amounts. Treat those three months as your calibration window.

Admins also get enterprise/cost-center/user budget controls. A $0 user budget means no access at all, which will be the way to curb runaway costs if you have developers who just can't help using the more expensive models for every single task.

But don't be surprised if they then try to expense their $200 USD monthly Claude Code subscription instead. 😅

What to do now

1. If you're annual: Check renewal date and model usage now. Multiplier changes land June 1.

2. If you're agent-heavy: Audit model choice immediately. Save frontier models for hard tasks.

3. If you're org/admin: Configure budgets before June 1. Pooled credits are great until one developer drains the whole pool on day six of the month.

4. For everyone: Build lightweight-first habits. GPT-4.1/GPT-5 mini for routine work, heavier models for genuinely hard problems.

5. Use the preview bill as soon as it appears. If you skip this, don't act surprised later.

The bigger picture

None of this is shocking if you've watched the economics. I've been writing about the end of freebie AI, and this is exactly that playbook: agentic workloads don't fit flat-rate pricing forever.

As GitHub put it, "it's now common for a handful of requests to incur costs that exceed the plan price." And while this change stings, it's understandable from a business perspective.

For developers not running constant heavy agent sessions, impact may be moderate. For annual subscribers using expensive models, impact can be very real.

Headline prices stay the same, completions stay unlimited, and everything else gets metered.

And as a small silver lining, it looks like GitHub isn't going to bill for a full "AI Credit" for small interactions that cost less than $0.01. So if you have a 0.5 credit interaction, you won't be charged 1 full credit — you'll be charged the actual cost, which is surprisingly more lenient than GitHub's pricing with GitHub Actions minutes, where you get charged for a full minute even if your job runs for 10 seconds.

Final thoughts

This has been a long time coming, but the shift of the whole industry towards usage-based billing models has been faster than I expected.

Plan accordingly. The buffet is closing — but the à la carte menu isn't unreasonable if you order carefully.

And if you're like me, your GitHub Business/Enterprise plan has looked like this for a while already anyway:

References

• GitHub Copilot is moving to usage-based billing (GitHub Blog, Apr 27, 2026)
• Models and pricing for GitHub Copilot (GitHub Docs)
• Usage-based billing for individuals (GitHub Docs)
• Usage-based billing for organizations and enterprises (GitHub Docs)
• Changes to GitHub Copilot Individual Plans (GitHub Blog, Apr 20, 2026)