GitHub Copilot's AI Credits: What June 1 Means for Your Monthly Bill

On April 27, GitHub announced that Copilot is moving every plan to usage-based billing on June 1, 2026. The old premium request units (PRUs) are being replaced by GitHub AI Credits — a token-based system where 1 credit equals $0.01 USD. Your monthly subscription now buys a credit balance, and you spend those credits based on the model and the actual input, output, and cached tokens your sessions consume.

If you're on a monthly Copilot Pro or Pro+ plan, you migrate automatically. Annual subscribers keep the old PRU pricing until their plan expires — but with higher model multipliers starting June 1, so the same workload costs more. There's about a week left to set rate-limit alerts before the switch.

How Many Credits You Get

The included credit amount matches your subscription price one-for-one:

Copilot Pro       $10/mo   →  $10 in AI Credits
Copilot Pro+      $39/mo   →  $39 in AI Credits
Copilot Business  $19/seat →  $19 in AI Credits
Copilot Enterprise $39/seat →  $39 in AI Credits

# Promotional bonus, June-August only:
Business          $19/seat →  +$11 = $30 effective
Enterprise        $39/seat →  +$31 = $70 effective

Once you exhaust your monthly credits, additional usage is billed at the same $0.01-per-credit rate. You can set a hard spend cap in the billing dashboard — turn this on now if you haven't. Default behavior is to keep charging until you cancel.

What Still Costs You Nothing

Two features stay outside the credit system. Code completions in your editor — the gray ghost-text suggestions — remain unlimited on every plan. So do Next Edit Suggestions, the multi-line predictions that fire as you save a file. These are the highest-frequency Copilot features for most developers, so the included credits will stretch further than the headline number suggests.

What Burns Credits

Everything that involves a chat-style call now consumes credits. That includes Copilot Chat (in any IDE), Copilot CLI, the cloud-hosted Copilot agent that runs in your browser, Copilot Spaces, Spark, and any third-party coding agent that bills through GitHub. Code review on PRs also draws from credits — and it stacks GitHub Actions minutes on top, since the review runs as a workflow.

Token cost varies by model. A Sonnet 4.6 chat turn might consume 5-15 credits. A GPT-5 Codex run for a long task can spike into the hundreds. Opus 4.7, which is now Pro+ only, is the most expensive per token. The CLI now prints credit cost after each turn so you can watch the meter in real time.

Monthly vs. Annual Subscribers

If you signed up monthly, you have no choice — June 1 flips you to credits. If you signed up annually, you stay on the old PRU model until renewal. But the trade is that model multipliers go up for annual customers on June 1, so a Claude Sonnet call that used 1 PRU before might cost 1.5 PRUs after. The numbers are designed to make annual and monthly cost similar over a full year.

What to Do This Week

Three actions before June 1. First, set a spend cap — anything from $0 (no overages allowed) to a few hundred is fine, but explicit is better than default. Second, audit which models you're calling. The model picker in Copilot Chat now displays the credit cost per turn next to each model name; pick the cheapest one that works for the task. Third, if you run Copilot agents in CI, set an environment variable to cap their per-job spend so a runaway agent can't drain a month of credits in one PR.

# Check your current usage
gh copilot usage --month

# Set a hard monthly cap (in dollars)
gh copilot billing set-cap --amount 50

# Cap per-job spend for CI agents
export COPILOT_MAX_CREDITS_PER_JOB=20