GitHub Copilot’s Credit Cushion Ended on September 1

GitHub Copilot’s extra Business and Enterprise credits ended on September 1, 2026, three months after usage-based billing went live.

Teams now get 1,900 AI credits per Business seat and 3,900 per Enterprise seat. Unused credits still vanish at month’s end, and a cheaper model no longer takes over when the pool hits zero.

The Switch From Request Units to AI Credits

Mario Rodriguez, GitHub’s chief product officer, set out the move to usage-based billing on April 27, 2026. Premium request units gave way to GitHub AI Credits on June 1. Seat prices did not move. What those prices buy did.

1 AI credit equals $0.01. Chat, the CLI, the cloud agent, Spaces, Spark, and third-party coding agents draw from the pool. Each turn is priced from the model you pick and the tokens it reads, writes, and caches, then converted into credits at published API rates.

Today, a quick chat question and a multi-hour autonomous coding session can cost the user the same amount. GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable.

Mario Rodriguez, Chief Product Officer, GitHub Blog

Rodriguez’s point was that Copilot had become an agent platform, and a flat request no longer matched the compute. GitHub’s own account posted the June 1 date the same day as the blog.

Starting June 1st, GitHub Copilot will move to a usage-based billing model as GitHub Copilot supports more agentic and advanced workflows.

In early May, you'll see a preview bill experience, giving visibility into projected costs before the transition.

👉 Read more about the…

— GitHub (@github) April 27, 2026

The GitHub Community thread that followed, discussion 192948, ran to 534 comments. The complaint that stuck was not the sticker on the plan card. It was that a prompt now had a price, and that price was hard to see before you sent it.

3,000 Business Credits Became 1,900 on September 1

Individual plans keep the April dollar price and add a second bucket on top. GitHub documents base credits plus a flex allotment for Pro, Pro+, and Max. Base credits match the monthly fee and do not change. Flex is extra included usage that GitHub says it can move as model prices and efficiency change. Base is spent first. Flex then applies on its own across the IDE, GitHub.com, and the CLI.

INCLUDED AI CREDITS BY PLAN

That flex bump is why a $10 Pro plan now includes $15 of model usage, and why Max includes $200 of usage for a $100 fee. Organization plans never got that split. They got a time-limited extra pool instead.

Existing Business and Enterprise customers received promotional credits through September 1. Business sat at 3,000 credits per user per month. Enterprise sat at 7,000. After that date the included amounts returned to 1,900 and 3,900. That is 1,100 fewer credits on Business and 3,100 fewer on Enterprise.

Credits are pooled at the billing entity, not locked in per-person buckets. GitHub’s own example is an enterprise with 100 Copilot Business users sharing 190,000 credits. During the promo that same seat count shared 300,000, which is 110,000 more. Light users still offset heavy ones. The whole pool is just smaller.

Included credits do not carry over. Unused amounts are forfeited, and the allowance resets on the first day of each calendar month. For a team that treated June through August as the new normal, September is the first month on the numbers GitHub called standard in April.

What a Frontier Model Session Costs

The meter is not one price. GitHub publishes per-1-million-token rates by model, and the spread is wide enough that model choice is a budget decision, not a preference.

SAMPLE COPILOT MODEL RATES PER 1M TOKENS

Claude Sonnet 4.6 input costs 15 times GPT-5.4 nano. GPT-6 Astra input costs 50 times the same lightweight model. Agent mode and the cloud agent can fire several of those calls inside one task, and a long chat carries the earlier turns back in as input. That is how a “session” becomes a bill.

Gemini 3.8 Flash is cheaper while the promo lasts, at $0.75 input and $3.75 output through December 31, 2026. GitHub also says paid users get a break when they let Copilot pick the model, but the expensive path is still the one people open for hard work.

In the first days after June 1, one developer on the $39 Copilot Pro+ plan said about 8% of the monthly allotment went in two hours, and guessed the 7,000-credit quota might not last two days. Another said a single change request cost more than $6 and called the drain “impossible to predict.” A Reddit user said a Claude 4.8 pass to fix website issues used 1,180 credits, about 16% of a Pro+ month, for results they called only mediocre. 1,180 credits is $11.80.

Those are user claims, not GitHub’s ledger. They match the design. A frontier model, a fat context window, and an agent loop are the expensive combination, and the old request unit hid that mix.

No Cheaper Model When the Pool Hits Zero

Under premium request units, exhausting the quota often dropped you onto a cheaper model and you kept typing. That fallback is gone. When credits run out, work stops unless you upgrade, set a budget for extra usage, or wait for the next monthly reset.

For individuals, extra usage is a dollar budget that draws down at the same $0.01 per credit. GitHub may cap how much extra you can buy, based on usage patterns, billing history, and verification, and it pushes heavy users toward the next plan. An upgrade charges only the price difference for the rest of the cycle, and earlier usage counts against the new, larger allowance.

For organizations, extra usage is on by default. An admin who wants a hard stop has to turn off the AI credits paid usage policy. User-level budgets can halt one person even when the shared pool still has credits. A $0 user budget blocks that person immediately. There is no automatic step-down to a cheaper model when any of those caps hit.

Copilot code review adds a second meter. Token use still takes AI credits, and the agentic run also consumes GitHub Actions minutes at the same per-minute rates as other workflows. The model used for those reviews is chosen automatically and is not shown, so the credit hit is harder to estimate than a chat turn you picked yourself.

Paid Plans Still Include Unlimited Completions

The one Copilot habit GitHub did not put on the clock is the one that lives in the editor. Code completions and next edit suggestions are not billed in AI credits. They stay unlimited on every paid plan.

That split is easy to miss if you only read the June complaint threads. Autocomplete can still feel like the old Copilot. Chat, agents, review, and the CLI are the parts that now look like an API bill. People who live in inline suggestions can stay on Pro. People who run long agent sessions across a repo are the ones who hit 1,900 or 7,000 credits first.

The practical effect is a two-layer product. GitHub kept the free-feeling completions that trained the daily reflex. It priced the agent work at the same kind of token rates you would pay outside the wrapper. Once that split is visible, the wrapper is optional for the expensive layer.

Overflow Work Moved to Direct Model Access

The June threads filled with exit plans, and the ones that lasted were routing plans, not full cancellations. Completions stay in Copilot. The heavy turns go somewhere the balance does not reset on the first of the month.

WHERE THE EXTRA TURNS WENT

• Direct APIs: Anthropic and OpenAI accounts billed at the provider’s own token rates, with no GitHub credit wrapper on top.
• OpenRouter: One key, many models, no monthly forfeit; the company says it may expire unused credits after one year.
• Editor add-ons: RooCode, LM Studio, and custom VS Code endpoints that point Copilot-style chat at cheaper models such as DeepSeek, Kimi, and Qwen.
• Hybrid months: Spend the Pro+ pool first, then send the rest of the month through OpenRouter inside the same editor.

OpenRouter’s FAQ is the contrast that matters for anyone who hated watching credits hit zero on the first. The company says it may expire unused credits after one year of purchase. That is not a promise they last a year. It is also not a calendar-month wipe. A balance you do not spend in September can still be there in October, which GitHub’s included credits cannot.

Some teams tried to have both, bringing their own key into Copilot chat. At least one June report said a BYOK path through OpenRouter still deducted Copilot credits as well as the provider charge. Treat that as a configuration trap until your own usage dashboard shows a single bill. The cleaner split is the one users described in June: Copilot for the included pool, then a separate router when that pool is gone.

The quieter change is habit. A prompt that used to feel free now gets a model pick, a context trim, and a question about whether it belongs on this month’s GitHub pool at all. That shopping reflex is what the meter taught, and it does not go away if GitHub later adds a few more flex credits.

Copilot Max and the October Seat Charge

GitHub’s answer to the drain was not a return to flat requests. It was a larger pool, a dashboard, and tighter payment rails.

On June 1 the company turned the meter on for every plan and enabled Copilot Max as an upgrade for existing Student, Pro, and Pro+ subscribers. Max is $100 a month for 20,000 credits. New individual sign-ups were paused that week, with a promise to reopen later. User-level budgets shipped in the same drop, so a runaway agent can be stopped on purpose instead of by surprise.

HOW THE METER ARRIVED

1. April 27, 2026: Rodriguez announces the June 1 switch, unchanged seat prices, and a three-month credit bump for existing Business and Enterprise customers.
2. May 12, 2026: GitHub adds flex allotments for Pro and Pro+ and names the $100 Max plan, lifting included individual usage above the raw monthly fee.
3. June 1, 2026: Usage-based billing goes live. Code review starts consuming Actions minutes as well as AI credits. Max upgrades open. Some new sign-ups stay paused.
4. September 1, 2026: The Business and Enterprise promo ends. Included credits return to 1,900 and 3,900 per user.
5. October 1, 2026: Existing Copilot Business and Enterprise customers who pay by credit card or PayPal take an upfront charge on assigned seats, and extra usage may need another payment before work continues.

The September 3 changelog that reopened Business and Enterprise sign-ups also tightened those card and PayPal paths. New seats in that group require payment before Copilot access. Included usage may be prorated across the month to match seat proration. Plan prices, GitHub said, are not changing.

Spending limits and the usage dashboard are still the tools GitHub points at. They help you see the burn. They do not put unused credits into the next month, and they do not bring back the cheap-model fallback. For a Business team that just lost 1,100 credits per seat, the next decision is whether 1,900 is a ceiling, a starter balance, or a reason to keep only completions on GitHub and send the agent work somewhere else.

October 1, 2026 is when card-billed organization seats start paying up front, and when extra usage can demand another payment before the agents run again.

Frequently Asked Questions

How many AI credits does GitHub Copilot include each month?

Base credits on Pro, Pro+, and Max match the monthly price and GitHub says those base amounts never change, while the flex allotment is written to move as model prices and efficiency change. Copilot Free and Copilot Student also get a credit allowance, but they reach models through auto model selection only.

Do unused GitHub Copilot AI credits roll over?

No. The pool resets at 00:00:00 UTC on the first day of each calendar month, and that clock does not follow your subscription billing date. Adding organization licenses mid-cycle grows the shared pool at once, but removing licenses does not shrink it until the next cycle starts.

Does GitHub Copilot charge AI credits for code completions?

Not on paid plans, where completions and next edit suggestions stay unlimited. Copilot Free is capped at 2,000 completions a month, and Copilot Student gets unlimited completions. Code review is separate: it still draws AI credits and also consumes GitHub Actions minutes.

What happens when GitHub Copilot AI credits run out?

You can upgrade, set a dollar budget for extra usage, or wait for the monthly reset, but extra usage may be capped, and GitHub tells heavy individual users to move up a plan rather than keep topping up. Plans bought through GitHub Mobile on iOS or Android cannot purchase extra credits at all. In organizations, extra usage stays on unless an admin disables the AI credits paid usage policy.

What is a GitHub AI credit worth?

One credit is $0.01, so a $10 extra-usage budget covers 1,000 credits. Paid users also get a 10% discount on model costs when they use auto model selection in Copilot Chat, Copilot CLI, the Copilot app, or Copilot cloud agent. Annual Pro and Pro+ subscribers stay on the older request-based multipliers until those annual plans expire, then they move to monthly usage-based billing or to Copilot Free.