GitHub Copilot Code Review used to be included, from June 1st you pay twice

Home All Posts GitHub Copilot Code Review used to be included, from June 1st you pay twice

Trends,

AI in Software Engineering

29/05/2026

GitHub Copilot Code Review used to be included, from June 1st you pay twice

Kendrick Curtis

13 mins read

In this article:

Subscribe to our blog:

Share:

-->

As LLMs mature and the model providers seek a return on their investment following huge spending on datacentres, it’s inevitable that the price of service for users was going to increase. We’ve already seen Anthropic try to remove Claude Code from their basic Pro plan because the traditional subscription model of “most people won’t max this out” (which used to work when there was a human in the loop) breaks down when agents are increasingly working autonomously or being driven by Ralph Loops.

Now it’s GitHub’s turn. Instead of a subscription that allows use of a set of features, the subscription becomes only a token allocation. Let’s take a look.

What's changing

Starting June 1, 2026, GitHub Copilot replaces premium request units (PRUs) with GitHub AI Credits. Usage is metered by token consumption, covering input, output, and cached tokens at published per-model rates, where one AI credit equals $0.01.

Credits come from a single pool, and almost every Copilot feature draws from it: Copilot Chat, the Copilot CLI, the cloud agent, Spaces, Spark, third-party coding agents, and Copilot code review. There is no per-feature budget. Code review pulls from the same credits as everything else, and the pool is shared across your whole organization rather than split into per-seat buckets. On the one hand, this allows more flexibility, so that under-users can have their spare tokens used up. However, this means that, when you run out across the org, almost everything stops.

Only the most basic human-interactive stuff remains unlimited (code completions and Next Edit Suggestions) even when the credit pool is depleted.

For Copilot Code Review in particular, there is a sting in the tail: token consumption is billed in AI Credits, and the agentic infrastructure that runs the review consumes GitHub Actions minutes, charged for private repositories at standard Actions rates. This is the same as their other Action-triggered agent pricing. These are separate line items. Actions minutes do not draw down AI Credits, and AI Credits do not cover Actions minutes. Both show up on the same bill.

For an existing Copilot Business customer, the included allotment is 3,000 AI credits per user per month from June 1 to September 1, 2026. That is a promotional rate; it drops to 1,900 credits per user per month after September 1.

One more change. When the pool and any overage budget are spent, there is no fallback to a cheaper model . Under PRUs, Copilot dropped to a lower-cost model once the budget ran out. That behavior is being removed. Once credits are gone, chat, agents, and code review stop entirely until the entitlement resets, or you buy more.

Hnnnnnnnnnnng

This sort of hard-stop usage-based billing can be, of course, infuriating! In a large organization where additional cost approval is not straightforward, that could mean the entire department locking down for days until the approval is acquired and the limits cranked.

Additionally, token-based billing while the norm with AI products is entirely unpredictable. If the CTO (hey that’s me!) decides to go on a weekend bender of multi-agent programming, suddenly everyone stops unless you put your hand in your pocket. Or you need to go set usage limits per-individual into GitHub, which is essentially the same system as before for pricing but now you have to manage it yourself.

For GitHub Copilot Code Review specifically, the cost of the review is unknowable in advance, since it’s governed by both input and output tokens, and also how much time the model thinks for is also both uncontrollable and billed to you. The model in use is not disclosed, so you can’t even do some finger-in-the-air maths on it.

The shared pool makes this worse. Code review has no protected budget. It competes for the same org-wide credits as every chat session, every agent run, and every CLI invocation. A few engineers running heavy agentic sessions on frontier models can draw down credits that would otherwise have covered code reviews for the rest of the team. With no cheap-model fallback, a pool that empties mid-month means code review stops until the next cycle. Usage-based billing has not started yet, so this is a forward-looking risk rather than a measured outcome. It is a structural one, written into how the pool works.

The reaction from developers has been skeptical. GitHub's own announcement drew 37 downvote reactions and a handful of "confused" reactions, with almost no positive ones. Visual Studio Magazine summarized the mood in a headline: "You Will Get Less, but Pay the Same Price." IT Pro noted that GitHub's FAQ...