GitHub Copilot premium requests are now AI credits

In brief

GitHub Copilot premium requests ended on 1 June 2026 for every plan except annual Pro and Pro+, as of November 2026. GitHub now meters usage in AI credits worth $0.01 each. Copilot Pro at $10 a month includes 1,500 credits and Pro+ at $39 includes 7,000. Completions never use credits.

Are GitHub Copilot premium requests still a thing? No, they are AI credits now

Premium requests are no longer GitHub's billing unit for any monthly Copilot plan. GitHub's 27 April changelog says "Starting June 1, premium request units (PRUs) will be replaced by GitHub AI Credits." The old name still gets searched, so it's worth being exact. Here's the old meter next to the new one, from GitHub's own pages read on 14 November 2026:

Two black plinths in a dark studio: a worn brass tally clicker with a blank face on a faded paper ticket, and a glowing glass fuel gauge with an amber needle, a code editor window behind them.
Premium requests (legacy)AI credits (current)
UnitOne request, times a model multiplierTokens, priced per model, 1 credit = $0.01
Applies toAnnual Copilot Pro and Pro+ plans that were not moved on 1 June 2026Every other Copilot plan
Out of allowanceSet a budget for more, or upgradeBuy more, upgrade, or wait; no fallback model
Reset00:00:00 UTC on the 1st of each month00:00:00 UTC on the 1st of each month
CompletionsIncludedIncluded, never use credits

GitHub's reason was blunt. Its announcement post says "the current premium request model is no longer sustainable" and that Copilot "is not the same product it was a year ago." The old meter counted requests. The new one counts tokens.

If you search the word "premium" in GitHub's current docs, you'll find it only on pages that are marked legacy. The page that used to explain the meter now opens with "This article only applies to Copilot Pro and Copilot Pro+ subscribers on an existing annual plan." If you're on a monthly plan, the rest of this page is the live system. Annual plans come last.

How many AI credits does each GitHub Copilot plan get?

Copilot Pro gets 1,500 AI credits a month and Pro+ gets 7,000, as of November 2026. GitHub's plans docs split each allowance into base credits, which match the plan fee, and flex credits on top. Flex can change as model prices and efficiency change. Your base credits are spent first, and then the flex credits apply on their own.

Plan, as of November 2026PriceBase creditsFlex creditsTotal a monthDollar value
Copilot Free$0Not printedNot printedAn allowance, no numberNot printed
Copilot Student$0, verified studentsNot printedNot printedAn allowance, no numberNot printed
Copilot Pro$10 a month1,0005001,500$15
Copilot Pro+$39 a month3,9003,1007,000$70
Copilot Max$100 a month10,00010,00020,000$200
Copilot Business$19 per seat a monthNot split outNot split out1,900 per user$19
Copilot Enterprise$39 per seat a monthNot split outNot split out3,900 per user$39

The dollar column is arithmetic from GitHub's "1 AI credits = $0.01 USD". The April changelog described Pro as "$10 in monthly AI Credits", and the docs list 1,500. The two agree once you split it: $10 is the 1,000 base credits, and the other 500 are flex credits on top. GitHub prints no number for Free or Student. Copilot Free keeps its separate cap of 2,000 completions a month, covered in is GitHub Copilot free.

What do 1,500 credits buy? That depends on the model, because GitHub's pricing table charges per million tokens. I priced one reference turn on that table: 20,000 input tokens and 2,000 output tokens, with no caching. Think of it as a chat with a medium sized file attached to it. It's my arithmetic, not a GitHub figure, and real sessions cache and vary.

ModelInput and output per 1M tokensCredits for the reference turnTurns on Pro's 1,500
Claude Haiku 5.5$0.10 and $0.500.35,000
Claude Sonnet 5.5$2.00 and $10.006250
GPT-6 Sol$2.00 and $10.006250
Claude Opus 5.5$4.00 and $20.0012125
Claude Fable 5.2$8.00 and $40.002462
GPT-6 Astra$10.00 and $50.003050
One Pro allowance buys 5,000 Haiku 5.5 turns or 50 GPT-6 Astra turns
505005,000GPT-6 AstraClaude Fable 5.2Claude Opus 5.5Claude Sonnet 5.5GPT-6 SolClaude Haiku 5.5

The spread is 100 to 1 on the same Pro allowance. Long threads and agent runs sit well above the reference turn, since every turn re-sends the history.

When do Copilot AI credits reset, and what counts against them?

Copilot credits reset at 00:00:00 UTC on the first day of each calendar month, and unused credits don't carry over. GitHub's billing docs say the date doesn't move with your billing cycle. If you sign up on the 28th, you get a fresh allowance three days later. The old premium request counter used the same clock, so nothing changed there.

Credits are spent on Copilot Chat, Copilot CLI, the cloud agent, Spaces, Spark and third party coding agents. Code completions and next edit suggestions are not billed in credits and stay unlimited on paid plans. GitHub names conversation length, agentic features such as agent mode, and model choice as the main drivers of what a session costs.

Two discounts exist as of November 2026. Paid plans get 10% off model costs when auto model selection handles Copilot Chat or the cloud agent. Through 30 November 2026 paid plans also get 30% off when using Copilot Fusion. Neither of them is a permanent rate, so don't build a budget on the second one. Plan for the full price.

Business and Enterprise pool credits. GitHub's organization billing docs say a user's included credits "are pooled at the billing entity level", so 100 Business seats share 190,000 credits. The April changelog promised $30 per Business seat and $70 per Enterprise seat in credits for June, July and August, which means that offer is over.

What happens when you run out of Copilot AI credits?

On an individual plan you can buy more, upgrade, or wait for the first of the month, and there is no cheaper model to fall back on, so the work simply stops until you pick one of those three. The April changelog says "Fallback experiences will no longer be available." Under premium requests, GitHub let you drop to a lower cost model when the allowance ran out. That net is gone.

GitHub's billing docs list the choices when your allowance is used up. Each one costs something different:

ChoiceWhat it doesWhat it costs
Upgrade to the next tierRaises your allowance at onceOnly the price difference, and earlier usage counts toward the bigger allowance
Set an additional usage budgetLets you keep working past the allowance$0.01 a credit, so a $10 budget covers 1,000 credits; usage may be capped, and you pay for what you already used before continuing
Wait for the resetFresh allowance at 00:00 UTC on the 1stNothing

Users in an organization get a banner that asks them to request more budget from their admin. If additional usage isn't allowed there, the docs say "Usage is blocked until the next billing cycle when monthly amounts are refreshed."

Whizi, which publishes this page, can't replace Copilot in your editor. It has no plugin for one and doesn't run Copilot. If it's the chat questions about code that you keep running out of, a free Whizi account gives you 10 messages a day on Whizi AI, Auto and Claude Haiku 5.5. It also gives you 2 pictures a month and 5 photo uploads a day, with no card. Whizi Pro carries Claude, GPT and Gemini, and its first week is $1.99. Use Copilot for completions and Whizi for the long questions that burn credits.

If the wall you hit says daily limit instead, that's a different product. Microsoft Copilot daily limit reached covers it.

Which Copilot plans still use premium requests?

Only Copilot Pro and Pro+ subscribers on an existing annual plan, until that plan expires. GitHub's April changelog says they "will remain on their existing plan" with premium request pricing, and that their model multipliers rose on 1 June. At expiry they move to Copilot Free unless they upgrade to a paid monthly plan. Switching to monthly early gets prorated credits for the rest of the annual term, per GitHub's announcement post.

GitHub's legacy docs print a multiplier for five models. A multiplier is how many premium requests one use of the model consumes:

ModelPremium requests per use (annual plans)
MAI-Code-1.1-Flash0.25
Claude Haiku 4.50.33
GPT-5.3-Codex6
Claude Sonnet 4.69
Claude Opus 4.827

The legacy pages print no allowance and no overage price. They say only that "Allowances vary by plan." The one place the old numbers survive is a third party. Xebia, in a blog post updated 22 June 2026, reported 300 premium requests a month for Pro, 1,500 for Pro+, and $0.04 per extra request. Treat those as Xebia's report, not GitHub's. If they hold, 300 requests is 11 uses of Claude Opus 4.8 at a multiplier of 27, or 33 uses of Claude Sonnet 4.6 at 9.

Check which meter you're on before you plan anything, since the two meters count different things. If you're weighing an editor with another pricing model, Cursor vs Copilot compares the two.

Which Copilot plan is enough for you?

Pick by how much agent work you do, not by how many chat messages you send, because tokens are the meter. If you only ask a few questions in a day, you won't notice the meter at all. It's the people who run agent mode or the cloud agent every day that will feel it.

You arePickWhy
Mostly using completionsCopilot Free at $0Completions don't use credits; 2,000 a month
A verified studentCopilot StudentFree, with unlimited completions and an allowance of credits
Chatting and editing with a mid-size modelCopilot Pro at $101,500 credits, unlimited completions
Running agents most of the dayPro+ at $397,000 credits, a higher allowance than Pro
Heavy agent use on top modelsCopilot Max at $10020,000 credits
On an annual Pro or Pro+ planStay put to expiryPremium requests continue; check the multipliers

Students have a path of their own, covered in GitHub Copilot student discount. And if you're ending a plan instead, how to cancel GitHub Copilot walks through it.

Workflow checklist
  • Check whether your account is on AI credits or the legacy annual premium request meter
  • Read your plan's monthly credits in GitHub's plans docs and convert at 1 credit = $0.01
  • Use a light model for routine edits and keep heavy models for hard tasks
  • Open a fresh chat when a thread grows long
  • Set an additional usage budget only if you accept paying $0.01 a credit
Common questions

Frequently asked questions

What replaced premium requests in GitHub Copilot?

GitHub AI credits replaced premium requests on 1 June 2026 for every plan except annual Pro and Pro+. One credit is $0.01 and usage is charged by tokens at each model's own rate. Copilot Pro includes 1,500 credits a month, Pro+ includes 7,000, and Max includes 20,000, per GitHub's plans docs as of November 2026.

How many premium requests does GitHub Copilot Pro get?

Monthly Copilot Pro gets no premium requests now. It gets 1,500 AI credits for $10 a month. Annual Pro subscribers on the legacy meter still have premium requests, but GitHub's docs print no number for them. Xebia reported 300 a month for Pro in a post updated 22 June 2026, which is a third party report.

Do GitHub Copilot premium requests reset monthly?

Yes. Premium requests on legacy annual plans reset at 00:00:00 UTC on the first of each month. AI credits reset on the same clock, and neither carries over. The date follows the calendar, not your billing day. GitHub says the allowance resets to the full monthly amount on the first day of each calendar month.

Does GitHub Copilot Free get premium requests?

No. Copilot Free gets an allowance of AI credits that GitHub prints no number for, plus 2,000 code completions a month. Completions don't use credits. Xebia reported 50 premium requests a month for Free before the change, which is a third party figure that GitHub's current docs don't repeat.

What happens when you run out of GitHub Copilot AI credits?

You can buy more with a dollar budget, upgrade for the price difference, or wait for the first of the month. GitHub removed the fallback to a cheaper model that premium requests allowed. In an organization with extra usage turned off, usage is blocked until the next billing cycle. A $10 budget covers 1,000 credits.