GitHub Copilot no longer charges you per request. Since June 1, 2026, every GitHub Copilot plan has been billed in GitHub AI Credits, and each credit is worth exactly one cent. What a task costs now depends on two things, according to GitHub: the model you pick and the number of tokens it reads and writes. A quick question to a small model can cost a fraction of a credit, and a long agent session on a frontier model can burn through dozens. The monthly allowances are easy to remember: 1,500 credits on GitHub Copilot Pro, 7,000 on Pro+ and 20,000 on Max. The harder part is that the same allowance can last a month or an afternoon, depending almost entirely on which model you leave selected.

Everything below comes from GitHub Docs, the GitHub blog and the GitHub changelog, checked on October 9, 2026, Beijing time. GitHub's prices are stated in US dollars.

What replaced premium requests

Before June 1, GitHub Copilot counted premium requests. Each model interaction cost one premium request unit, and a multiplier scaled that up or down by model, so more powerful models used more premium requests. When you ran out, GitHub Copilot could drop you to a cheaper model so you kept working.

GitHub announced the change on April 27. Its chief product officer, Mario Rodriguez, wrote that "a quick chat question and a multi-hour autonomous coding session can cost the user the same amount," and that "the current premium request model is no longer sustainable." The replacement bills every interaction by its input, output and cached tokens at each model's published rate, then converts the total into AI Credits. The fallback went with it. GitHub said fallback experiences "will no longer be available," and its docs for organizations now state plainly that there is no automatic fallback to lower-cost models when a budget runs out.

Premium requests haven't disappeared for everyone, though. If you were on an annual GitHub Copilot Pro or Pro+ plan, you stayed on request-based billing until that plan ends, and GitHub raised the multipliers for those annual subscribers on June 1. The legacy table now lists GPT-5.5 at 57 premium requests per interaction, Claude Opus 4.8 at 27, Claude Sonnet 4.6 at 9 and GPT-5 mini at 0.33, while each GitHub Copilot code review costs 13. GitHub also says users on legacy annual plans won't get new models and features. When the annual plan expires, you're moved to GitHub Copilot Free. You can switch early instead: cancel for a prorated refund, or upgrade to a monthly plan and receive prorated credit for what's left of the annual one.

If you see "multiplier" in an old guide, that's the giveaway. GitHub says multipliers belong only to the legacy system and don't apply to usage-based billing.

What each plan includes

GitHub's docs list these monthly allowances:

  • GitHub Copilot Free: an allowance of AI Credits, models through auto model selection only, and 2,000 code completions a month. GitHub doesn't publish the credit amount.
  • GitHub Copilot Student: free for verified students, also with an unspecified credit allowance and auto model selection only, but with unlimited code completions.
  • GitHub Copilot Pro ($10/month): 1,500 credits, made up of 1,000 base credits and a 500-credit flex allotment.
  • GitHub Copilot Pro+ ($39/month): 7,000 credits, 3,900 base and 3,100 flex.
  • GitHub Copilot Max ($100/month): 20,000 credits, 10,000 base and 10,000 flex.
  • GitHub Copilot Business ($19 per seat per month): 1,900 credits per user.
  • GitHub Copilot Enterprise ($39 per seat per month): 3,900 credits per user.

The split on the individual plans is deliberate. Base credits match the subscription price one for one, and GitHub says they never change. The flex allotment sits on top. GitHub describes it as "a variable part of your included usage" that is "designed to adapt as the economics of AI evolve, including model pricing, new models, and improvements in efficiency." The base is the part GitHub promises to keep; the flex comes with no such promise. Your base credits are spent first, and the flex kicks in automatically after that, at the same rates. That matters most on Max, where half of the 20,000 credits are flex.

This lineup didn't arrive all at once. GitHub's April announcement promised Pro $10 and Pro+ $39 in monthly credits, matching their prices. On May 12, before the switch, GitHub added the flex allotments and introduced Max, after users questioned whether the original amounts would "go far enough." Max opened on June 1 as an upgrade for existing Student, Pro and Pro+ subscribers, and GitHub began reopening new sign-ups for Student, Pro, Pro+ and Max on June 17.

Two things still don't cost credits at all. Code completions and next edit suggestions remain unlimited on every paid plan. Included credits don't roll over either: the allowance resets to the full amount at 00:00 UTC on the first day of each calendar month, which is 08:00 Beijing time, whatever your billing date.

Turning credits into dollars and tokens

The conversion is simpler than it looks. One AI Credit is $0.01, so 100 credits make a dollar, and the allowances translate directly: Pro's 1,500 credits are $15 of model usage, Pro+'s 7,000 are $70 and Max's 20,000 are $200. GitHub's own pricing page puts them in those dollar terms.

Converting to tokens takes one more step. GitHub publishes per-model prices in dollars per million tokens, split into input, cached input, output and, for many models, cache writes. Multiply any of those prices by 100 and you get credits per million tokens. A few current rates from GitHub's pricing table, for standard context:

  • Claude Haiku 5.5: $0.10 input and $0.50 output, so 10 and 50 credits per million tokens. Above 100,000 input tokens it rises to $0.50 and $2.50.
  • GPT-6 Luna: $0.10 input and $0.50 output, the same as Haiku 5.5.
  • GPT-6.1 Sol: $2 input and $10 output, or 200 and 1,000 credits per million.
  • Claude Sonnet 5.5: $2 input and $10 output.
  • Claude Opus 5.5: $4 input and $20 output.
  • GPT-6 Astra: $10 input and $50 output, or 1,000 and 5,000 credits per million. Above 272,000 input tokens it rises to $20 and $75.

GitHub says GitHub Copilot uses each model's listed API rates. On the October 7 launch of Claude Haiku 5.5, it said the model "is billed at provider list pricing."

GitHub has no calculator, so here is the arithmetic for a single request that sends 20,000 input tokens and gets 2,000 output tokens back, with no caching. On Claude Haiku 5.5 the input costs 0.2 credits and the output 0.1, so the request comes to 0.3 credits. On GPT-6.1 Sol the same request costs 6 credits. On Claude Opus 5.5 it's 12. On GPT-6 Astra it's 30, a hundred times the Haiku price. Put against Pro's 1,500 credits, that's about 5,000 such requests on Haiku 5.5 and 50 on Astra.

Real sessions won't be that tidy. An agent task makes many model calls, cached input is billed at a much lower rate than fresh input, and GitHub notes that switching models or changing reasoning effort mid-session invalidates the cache, so the whole context gets billed again as fresh input. The ratios still hold, and they are the point. On per-token prices alone, the model picker is now a hundredfold price dial. Auto model selection narrows the gap slightly in your favor, since paid plans get a 10% discount on usage billed through it.

One feature resists this math. GitHub says GitHub Copilot code review picks its model automatically and doesn't disclose it, so per-review costs vary and you can't predict them from the table. Since June 1, each review on a private repository also consumes GitHub Actions minutes, billed at standard Actions rates once you're past your included minutes. Reviews on public repositories still use free Actions minutes.

When the credits run out

On an individual plan, GitHub gives you three options when the included credits are gone. You can upgrade, paying only the price difference, and GitHub says the new plan's larger allowance is available immediately, minus what you've already used this month. You can stay put and set a dollar budget for additional usage, which draws down at $0.01 per credit, so a $10 budget covers 1,000 credits. Or you can wait for the reset on the 1st.

The budget route comes with a catch. GitHub's June 1 changelog says it "may limit your total additional AI Credits based on your usage patterns, billing history, and verification status," and the docs warn that additional usage "may be capped." GitHub doesn't say where that cap sits. If you hit it, you have to pay for the additional usage you've already consumed before you can carry on. If you subscribed through GitHub Mobile on iOS or Android, buying additional credits isn't available at all. GitHub can email you when included usage reaches 90% and 100%, and budgets you set yourself can alert at 75%, 90% and 100%.

For long agent runs, there's a finer-grained control. In GitHub Copilot CLI and the GitHub Copilot SDK you can set an AI Credit session limit before a task starts, and the agent stops cleanly when it reaches it and asks whether to continue. GitHub describes these as soft limits that don't replace budgets.

How GitHub Copilot Business and Enterprise differ

Organizations get the same per-token pricing with different plumbing. Each Business seat adds 1,900 credits and each Enterprise seat 3,900 to a shared pool at the billing-entity level, so 100 Business users share 190,000 credits rather than 100 separate buckets. Adding seats mid-month grows the pool right away. Removing them only shrinks it from the next cycle. Existing Business and Enterprise customers got a promotional 3,000 and 7,000 credits per user for the first three months, June 1 to September 1, after which GitHub's docs say included usage returns to the standard amounts.

The default setting deserves an admin's attention. Additional usage is enabled by default for organizations and enterprises, so once the pool is empty, usage keeps going at the published rates unless an administrator turns off the AI Credits paid usage policy. From there, GitHub layers several budgets. A user-level budget caps how much one person can consume across both the pool and paid overage, and it always stops hard. Cost center, organization and enterprise budgets only apply after the pool is empty, and they stop usage only if "Stop usage when budget limit is reached" is switched on, which GitHub says is off by default. GitHub also warns that the enterprise budget isn't a total monthly budget. In its own example, 400 Business seats cost $7,600 in licence fees, and adding a $5,000 enterprise budget means a maximum bill of $12,600.

Two recent changes smooth the edges. Since September 16, members who hit their limit can request more budget, and owners can approve it from settings, though not in enterprises with managed users. And from October 1, Business and Enterprise customers who pay by credit card or PayPal are charged upfront for every assigned seat at the start of each billing cycle, and GitHub says included usage may be prorated across the month to match.

Beyond the bigger allowance, the plans differ mainly in models: GitHub's plan table gives Business "access to premium models" and Enterprise "priority access to premium models."

Credits reward knowing your models

The new system is fairer than premium requests in one obvious way: a one-line question no longer costs the same as an hour-long agent run. It also shifts the work onto you. The included numbers are generous at the bottom of the model range and thin at the top, and the flex part of each allowance can change without the price changing. If most of your GitHub Copilot use is chat and small edits, Pro's $15 of credits on a lightweight model will go a long way. If you live in agent mode on Astra or Opus-class models, expect to set a budget or move up a plan, and use session limits so one runaway task doesn't eat the month. OpenAI's Codex takes a different approach, with five-hour and weekly windows whose size OpenAI doesn't publish, which we covered in "OpenAI Keeps Codex's Weekly Caps Vague: What Each ChatGPT Plan Actually Gets."