GitHub Copilot goes metered June 1, 2026: three pay-once alternatives

On April 27, 2026, GitHub posted two announcements that landed on the same day. The first: Copilot's billing model moves entirely to usage-based AI Credits, with the changelog rolling out across all plans. The second: Copilot ревью кода starts consuming GitHub Actions minutes on June 1, 2026. Read together, they say one thing — Copilot is no longer a flat-rate product, and the meter starts running in 35 days.

The reaction was immediate. The GitHub Community discussion thread on the change crossed several hundred comments inside 24 hours. The Slashdot thread climbed past 200. Новое sign-ups for Copilot Pro and Pro+ have been paused since April 20, which means the people most exposed to the new pricing — small teams that used to live on the $10/seat plan — cannot even add seats during the transition.

This post does the math, names the two meters, and lists three pay-once alternatives that will not surprise you on June 2.

What actually changed

The old Copilot was simple: a flat seat price ($10/month for Pro, more for Pro+ and Business), and you could use the chat, code completion, and ревью PRer as much as you wanted. That model is being retired.

The new Copilot is metered along two dimensions:

1. AI Credits — every Copilot interaction now decrements a credit balance tied to your account. Heavy users (long agentic sessions, frequent ревью кода, multi-file edits) burn credits faster.
2. GitHub Actions minutes — starting June 1, 2026, Copilot ревью кода will count against the same Actions minute pool that runs your CI builds. Every ревью PR the bot performs is an Actions run.

The combined effect is the part that surprised people. From the GitHub community thread:

That sentence is doing a lot of work. Translated out of corporate-speak: the old plan prices were anchored to a usage pattern that no longer reflects how teams actually use Copilot. The fix is to charge per use. Which is fine for GitHub's revenue model and not fine for any team that planned a budget around the old per-seat number.

The math: a three-developer team, one year

Take a deliberately small team — three developers, one main repo, an average of 40 PRs per month between them. Under the old model, this is $30/month for three Pro seats, $360/year, predictable.

Under the new model, the same team carries:

• 3 base seats (still required for chat and completion) at the new metered floor.
• 40 ревью PRs/month against the Actions minute pool — every ревью that takes 90 seconds of compute is 1.5 minutes off the 2,000-minute free quota.
• AI Credit consumption that scales with how aggressively each developer uses agentic features.

The exact numbers will move as GitHub publishes final per-credit pricing, but every team that ran the calculation in the last 24 hours hit the same shape: predictable annual cost replaced by a usage curve that compounds whenever the team writes more code or ревьюs more PRs. Which is to say, whenever the team does its job.

The compounding-meter problem

Here is the part most write-ups have missed. Before June 1, a Copilot ревью PR was a Copilot operation. After June 1, it is a Copilot operation and an Actions run. Two meters burn at the same time for the same event.

For a team already running CI, the Actions minute side is not theoretical. The free tier on приватный репозиторийs is 2,000 minutes per month. A typical CI pipeline (lint + test + build) burns 4–7 minutes per push. Add an automated ревью PRer running on every push and you cross the 2,000-minute line earlier in the month than you used to. Past that line, Actions minutes cost real money.

What this means in practice: starting June 1, the cost of an active development month is no longer a function of the number of seats. It is a function of how much code the team writes. That is a fundamentally different model, and it is the one every other AI coding product (Cursor, Aider, Продолжить) has been moving toward for two years. GitHub is just the largest mover.

What "купите один раз" means in this context

Pay-once does not mean free. It means you trade a recurring meter for a one-time fee, and after that the only ongoing cost is the infrastructure you already pay for — your existing Actions minutes, your existing model API key.

For ревью PR specifically, the three pieces you need are:

1. A GitHub Action that runs on pull_request events.
2. A prompt template that produces useful ревью comments rather than nitpicks.
3. A model API key (Anthropic, OpenAI, or whichever provider you want) that you control directly.

The model is the only part that costs ongoing money, and it costs you the per-token rate the model provider charges, not a marketed-up Copilot reseller rate. For a 3-person team doing 40 PRs/month with Claude Sonnet 4.6 as the ревьюer, the API spend is tens of dollars per month, not hundreds.

The three alternatives

1. Septim Flint — $19 once

Sample wiring for the bug-spotter action looks like this:

name: PR Review — Bug Spotter
on:
  pull_request:
    types: [opened, synchronize]

jobs:
  ревью:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - name: Run bug spotter
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
        run: ./flint/bug-spotter.sh

Three lines of secret config, one shell script in the repo, no Copilot. The secret ANTHROPIC_API_KEY is yours; rotate or revoke it any time.

2. Septim Review Stack — $59 once

Review Stack is the answer when the question is "we want everything Copilot did plus the local pre-commit safety net." It includes Tether's three pre-commit hooks, which catch problems before they reach the PR — secret scanning, lint gate, and branch-name enforcement. That is one fewer Actions run per PR, which directly cuts your minute consumption.

3. Build it yourself with the Anthropic API directly

The minimal version is a single workflow file that fetches the PR diff, sends it to Claude with a ревью prompt, and posts the response as a comment. Работыing examples live in the БесплатноCodeCamp tutorial referenced in our ревью PR setup guide; we keep that post current as the API surface evolves.

What you do this week

1. Аудит your current Copilot spend. Pull the last three months of usage from your billing page. The pattern matters: if your team uses Copilot heavily for ревью кода (rather than just chat), you are the most exposed to the June 1 change.
2. Run the math at the new metered rate. GitHub will publish per-credit pricing before June 1. Multiply your team's ревью PR volume by the new credit cost. Сравнить to the flat $19 for Flint or $59 for Review Stack.
3. Pick a path before May 25. If you are migrating off Copilot, you want a working alternative in place before the meter switches over, not after. The June 1 date is also a convenient deadline for the team to learn the new wiring.
4. Decide on the local gate. Even if you keep Copilot for completion and chat, a pre-commit hook layer (Tether) cuts how often the cloud ревьюer runs. Less running = lower bill.

The longer view

Copilot is not the first AI tool to move to metered pricing, and it will not be the last. The pattern is consistent: a tool launches with a flat rate to grab market share, switches to metered once usage scales, and the bill grows whenever the team gets more productive. That dynamic is fine for shareholders and bad for budgets.

The pay-once playbook is the structural counter. You buy the tooling layer once, you control the model key, and the only cost that scales with your usage is the raw model spend — which you pay at provider rates без a markup.

Septim ships in this exact lane. Flint is $19. Review Stack is $59. Купите один раз. Keep it. The meter never starts.

Further reading

• Setting up Claude Code as your ревью PRer — the workflow guide if you want to wire your own Action.
• How to monitor Claude Code costs без surprises — the same discipline applied to the model side of the bill.
• GitHub: Copilot moves to usage-based billing — the official announcement.
• GitHub changelog: Copilot ревью кода will consume Actions minutes on June 1, 2026 — the changelog entry.
• The Register: Microsoft's GitHub shifts to metered — coverage of the changeover.