GitHub Copilot Pricing and GitHub Copilot Business Pricing After Premium Requests

The short answer: GitHub Copilot pricing for companies is $19 per user a month on Business and $39 on Enterprise, and premium requests no longer exist on those plans. On June 1, 2026 GitHub replaced them with GitHub AI Credits, billed on the tokens each interaction uses at the model's API rate, with 1 credit worth $0.01. Copilot Business still costs $19 per user and now includes 1,900 credits; Enterprise still costs $39 and includes 3,900. A premium request used to be a flat $0.04 times a model multiplier. A task now costs whatever its tokens cost.

That sounds like an accounting change. It is not. Under premium requests, a one-line chat question and a two-hour cloud agent session could cost the same single request. GitHub said so itself when it announced the switch, and called the old model no longer sustainable. If your team built its habits on agent mode during 2025 and early 2026, those habits now carry a real per-task price. For the full plan comparison and the Business vs Enterprise math, see our GitHub Copilot Enterprise pricing page.

GitHub Copilot premium requests vs AI credits, plan by plan

Seat prices did not move. What each seat includes did.

Three changes arrived with the switch and matter more than the numbers. Credits are now pooled across the company, so 100 Business seats share one 190,000 credit pool instead of 100 separate allowances. The fallback is gone: under premium requests, a user who ran out could keep working on an included model, and now Copilot simply stops when a budget runs out. And GitHub states that paid overage is enabled by default for organizations and enterprises.

What a Copilot task costs now, next to what it cost as a premium request

Under the old model, GitHub's own rules were simple. Copilot Chat used one premium request per prompt times the model multiplier. Copilot cloud agent used one premium request per session, plus one per steering comment. Tool calls the agent made on its own did not count. That made long autonomous sessions the best deal Copilot ever offered.

Under AI credits, the same work is priced by tokens. The token profiles below are our illustrative assumptions (a chat question at 20,000 input and 1,500 output tokens, an agent mode task at 300,000 fresh input, 1.2 million cached input and 40,000 output tokens, a long cloud agent session at five times that), priced at GitHub's published Copilot rates and leaving out cache write charges.

Read the bottom row twice. A session that cost one premium request, worth four cents at the overage price, now costs anywhere from under a dollar to fifteen dollars depending on the model. Chat got cheaper; autonomous work got much more expensive. That is the whole story of the change, and it is why teams that lean on cloud agent see the biggest jump.

Copilot code review changed too. It is now billed in AI credits for its tokens and in GitHub Actions minutes for the agent infrastructure it runs on, and GitHub picks the model automatically without disclosing it, so a review has no fixed price. GitHub's code review docs now give a range instead: $0.05 to $1 of AI credits per review at Lite effort and $0.25 to $5 at Balanced effort, before Actions minutes. On legacy annual individual plans, each review now uses 13 premium requests.

How to convert an old premium request budget into AI credits

There is no fixed exchange rate between the two units, because a premium request never measured tokens. What you can do is rebuild the budget from usage:

1. Pull one full month from the AI usage dashboard in your enterprise billing settings. September is the first clean month, because June to August ran on promotional pools.
2. Split usage by feature. Completions and next edit suggestions are free and unlimited on paid plans. Chat is cheap. Agent mode, cloud agent and code review carry almost all the cost.
3. Split usage by model. Between GPT-5 mini and a frontier model like Claude Fable 5.1, the cost of the same task differs by about 30 times. A model policy moves your bill more than any plan choice.
4. Compare the average per developer to 1,900 and 3,900. Below 3,900, Business plus overage is cheaper than Enterprise. At or above it, they cost the same, so Enterprise only makes sense for its features.
5. Set budgets before the next cycle, not after the invoice.

Copilot budget settings that replace the old request cap

Premium requests had a natural ceiling: when a user ran out, they dropped to an included model. AI credits have no ceiling unless you build one. GitHub gives you five controls:

One setting deserves thought before you roll it out. A user-level budget is a hard stop, and there is no fallback model anymore. Give an individual override, with an expiry date, to whoever holds the on-call rotation, so the engineer who gets paged for a production incident is not locked out of Copilot halfway through the fix. If you want zero spend above the included pool, an admin must switch off the AI credits paid usage policy in AI Controls; otherwise overage keeps billing.

Where a flat-priced agent is cheaper than Copilot credits

Credits are the right meter for open-ended coding work a developer drives. They are an expensive meter for jobs that repeat the same way every day. Reviewing every pull request against your conventions, turning bug reports into labeled and routed issues, and keeping API docs in step with the code are all repeated jobs. Run through Copilot on a frontier model, each one draws tokens from the pool every time.

A ready-made agent that does just one of those jobs runs on a flat monthly price with nothing to meter. The coding agents in our catalog, Code Reviewer, Bug Triager and Docs Writer, connect to GitHub and are priced per month, not per token, which makes them a predictable way to take the most repetitive volume off your Copilot pool and keep credits for the work that needs a developer at the keyboard. If you are weighing autonomous coding agents instead, compare Devin pricing and Windsurf seats and the Claude Code seats on our Claude Enterprise pricing page, and see how the metering models differ across vendors in our AI agent platforms comparison.

Who is still on premium requests

Only individual subscribers who paid for Copilot Pro or Pro+ annually before June 1, 2026. They stay on premium requests until the plan expires, with higher model multipliers from June 1 (Claude Sonnet 4.6 at 9 and GPT-5.5 at 57, for example), and they do not get new models or features. At expiry they move to Copilot Free unless they upgrade to a monthly paid plan, and GitHub offers prorated credit to anyone who converts early. For a company rollout none of this applies: every Business and Enterprise seat is on AI credits today.