The Short Answer

GitHub Copilot's subscription prices did not change on June 1, 2026. What changed is the unit of measurement: flat Premium Request Units (PRUs) are gone, replaced by GitHub AI Credits, where one credit equals $0.01. Instead of counting premium requests, every Copilot plan now includes a monthly allotment of GitHub AI Credits, with the option for paid plans to purchase additional usage. If you mostly use Copilot for inline code completion, the change is nearly invisible. If you rely on chat, agent mode, or the CLI, you are now on a live meter.

What Was the Old Model, and Why Did It Break?

The old Premium Request Unit system counted discrete requests, not tokens. A short request and a long one both consumed one unit, which made costs predictable but disconnected from the actual compute GitHub was paying for underneath.

The core driver behind the overhaul is the widespread adoption of Agent Mode. Under the PRU system, it cost the same number of points whether you asked for simple syntax help or had the AI spend minutes refactoring thousands of lines of code. But long-context conversations and cross-file agent operations consume massive amounts of tokens, and the legacy model simply could not shoulder the compute costs anymore.

In short, GitHub was subsidizing power users and needed to stop.

How AI Credits Work

Under AI Credits, multipliers no longer exist. One AI Credit equals $0.01 USD, and your cost is driven by actual token consumption: input tokens, output tokens, and cached tokens. Different models still cost different amounts, but the cost comes from token pricing rather than a multiplier applied to a request count.

Code completions remain unlimited and free; they do not touch the credit balance. The meter starts when you open chat, use agent mode, run Copilot CLI, or trigger code review.

Copilot bills against the input tokens, output tokens, and cached tokens each request consumes, priced at the published API rate for whichever model handled it. The conversion is straightforward: 1 AI Credit = $0.01 USD, so a model that costs $30 per million output tokens burns 3,000 credits for a million tokens of output.

What Each Plan Includes

Pro ($10/mo) includes $15 in credits (1,500 credits), Pro+ ($39/mo) includes $70 (7,000 credits), and Max ($100/mo) includes $200 (20,000 credits). For teams, Business ($19/user/mo) includes 1,900 credits per user and Enterprise ($39/user/mo) includes 3,900 credits per user.

The pattern is that the credit allowance is roughly dollar-for-dollar with the subscription price, except on Max, which was introduced alongside the billing change and carries a two-to-one credit-to-price ratio worth verifying directly with GitHub before you rely on it.

Existing Copilot Business and Enterprise customers are in a promotional period from June 1 through September 1, 2026, with higher included credit amounts than the standard allocations.

What Happens When Credits Run Out

By default, when your included credits run out, Copilot keeps working and bills overage with no cap. There is no automatic spending limit. GitHub added the control on July 2, 2026, but you have to turn it on manually: Settings → Billing → GitHub Copilot → Set spending limit. Setting it to $0 creates a hard stop when credits run out.

This is the biggest operational risk of the transition. Teams that did not set a limit before the first June billing cycle closed reported surprise overages.

Teams on Business and Enterprise plans also have AI Credit Pools: admins can cap how much of the shared monthly allowance any single user or cost center draws down.

Who Gets Hit Hardest

For a developer who uses Copilot the classic way, as tab-completion in the editor, nothing changed. For anyone leaning on chat, the CLI, or the coding agent, the bill is now a function of tokens, and that number swings by more than an order of magnitude depending on the task.

GitHub's own research published in May 2026 found that agentic tasks consume roughly 1,000x more tokens than a single-turn query. A single agent session writing a meaningful diff on a powerful model runs 600 to 1,200 credits ($6 to $12) in one interaction.

Heavy daily agent users are the ones most likely to exceed their included credits and pay overage, with total monthly spend reported around $60 to $100 for daily agent users.

A Pro user ($10/month, $15 in credits) doing mostly tab-completion plus occasional chat might spend $1 to $3 of credits per day. The $15 Pro allowance stretches a light-usage month comfortably, but a moderate-usage month will still tip into overage.

What About Annual Subscribers?

Not everyone is on the new system yet. Annual Pro or Pro+ subscribers ride out their current term on the old premium-request model. When their annual renewal comes up, they will be moved to credits. So if you bought an annual plan in October 2025, you are on premium requests until October 2026.

In June 2026, the multipliers for legacy annual subscribers also jumped sharply: Claude Opus-class models went to 27x (from 7.5x) and GPT-5-class to 6x (from 1x). That means annual subscribers on high-tier models may actually be paying more under the old system than they would under AI Credits, depending on their usage pattern.

How This Compares to Rivals

GitHub Copilot is not the only tool that moved this direction. Cursor moved to usage-based billing first, in June 2025. GitHub Copilot followed on June 1, 2026. The competitive context matters: Claude Code, Cursor, Windsurf, Replit, and Bolt.new all compete in this space, and most now use some form of token or credit billing for agentic workloads. The flat-fee advantage that distinguished Copilot for years is effectively gone.

Engineers doing heavy agent mode work are now paying API-equivalent rates, just routed through GitHub's billing. The math on Copilot Pro versus Claude Code or Cursor Pro is now worth running for your actual usage pattern.

Practical Steps to Take Now

Set a spending limit. Go to Settings → Billing → GitHub Copilot and set a monthly cap. This should be the first thing any individual or admin does before the next billing cycle.

Check which billing model you are on. If you are on an annual plan purchased before June 2026, you are still on PRUs until renewal. The experience and cost structure differs from a monthly subscriber on the same team.

Audit your agent usage. Agent mode bills by tokens, and agent loops consume tokens in a compounding way: the full context re-sends on every internal roundtrip. A task that feels short can be expensive. Use GitHub's billing overview page to see credit consumption before you run long sessions.

Match your plan to actual usage. The new model may be more economically rational for GitHub, but it is harder for users to predict. In developer tooling, predictability matters almost as much as raw price. If your work is completion-heavy, Pro still makes sense at $10. If you run multi-file agent tasks daily, do a realistic cost estimate against Pro+ or Max before assuming the base plan covers you.

The Bigger Picture

The PRU model was always an approximation. It made Copilot easy to budget but increasingly difficult for GitHub to sustain as agentic sessions grew in length and complexity. The move to AI Credits is a more honest representation of cost, and that honesty runs in both directions: users who do light work may find they stay well within their allowance, while power users now pay closer to what their usage actually costs to serve.

The real tradeoff is predictability. A flat monthly fee is easy to approve in a team budget. A metered bill that can spike with one long agent session requires more active monitoring. Setting spending limits and reviewing the billing dashboard regularly are now baseline habits for any team running Copilot at scale.