Pricing, checked against GitHub's own docs
GitHub Copilot Agent Pricing: What Agent Mode, the Coding Agent and Premium Requests Cost per Developer
The premium request is gone. Since June 1, 2026 every Copilot plan bills in AI credits at a cent each, which changed what an agent session costs and left most of the internet quoting an allowance that no longer exists. Here are the current numbers, and the two that decide your bill.
No card to start · Flat $29 a month, no usage meter
Pick a task
Plan
- planning
Files changed
Test run
Pull request
You review and merge. Agentcode never merges on its own.
The short answer
GitHub Copilot does not charge separately for agent mode or the coding agent. Both are included in every paid Copilot plan and both draw on the same meter: GitHub AI Credits, where 1 credit equals $0.01 of model usage at published per-million-token rates. Premium requests stopped being the billing unit on June 1, 2026, so any allowance quoted as "300 premium requests" is out of date. Included credits are $10 and 1,500 on Copilot Pro, $39 and 7,000 on Pro+, $100 and 20,000 on Max, $19 per seat and 1,900 on Business, and $39 per seat and 3,900 on Enterprise. Two things decide the real bill. Code completions are never billed, so agents are your entire variable cost. And the model you pick moves that cost about 17 times: the same agent task that costs 11 credits on GPT-5 mini costs roughly 188 on Claude Opus.
GitHub Copilot agent pricing, plan by plan
No Copilot plan sells the agent separately, so pricing the agent means pricing the plan and then pricing the usage underneath it. Every figure below is GitHub's published US list price and its published credit allowance, read from the plans and billing documentation in August 2026.
Swipe to see more
| Plan | Price | Base credits | Flex allotment | Total a month | What you get for agents |
|---|---|---|---|---|---|
| Copilot Free | $0 | Small monthly allowance | None | Not published | Limited agent access. Fine for trying agent mode on a personal repository, not for shipping work. |
| Copilot Pro | $10 / month | 1,000 credits | 500 credits | 1,500 credits | Agent mode in the IDE and the cloud agent, on a selection of models. The entry point for one developer. |
| Copilot Pro+ | $39 / month | 3,900 credits | 3,100 credits | 7,000 credits | Everything in Pro plus access to premium models, which is what agent work actually consumes. |
| Copilot Max | $100 / month | 10,000 credits | 10,000 credits | 20,000 credits | Priority access to premium models. Bought by people who kept exhausting Pro+ mid-month. |
| Copilot Business | $19 / seat / month | 1,900 credits per user | None published | 1,900 credits per user | An administrator has to turn the cloud agent on before anyone can use it. Credits pool across the enterprise. |
| Copilot Enterprise | $39 / seat / month | 3,900 credits per user | None published | 3,900 credits per user | GitHub Enterprise Cloud only. Same admin switch for the cloud agent, larger per-seat contribution to the pool. |
If you are choosing between the two organization tiers rather than pricing the agent, the seat arithmetic is on our GitHub Copilot Business page and the feature split is in Copilot Business against Copilot Enterprise. The $100 tier gets its own read in whether Copilot Max is worth it.
Premium requests became AI credits: what actually changed
Around 900 people a month still search for GitHub Copilot premium requests, and almost every page answering them quotes a table GitHub retired on June 1, 2026. The documentation describing premium requests is now marked legacy. Here is the before and after, row by row, with why each line changes a decision.
Swipe to see more
| The question | Premium requests, before | AI credits, now | Why it matters |
|---|---|---|---|
| What is the billing unit? | Premium requests | GitHub AI Credits, 1 credit = $0.01 | A request used to cost a flat multiple of one unit no matter how large it was. A credit is a cent of real model spend. The docs page describing premium requests is now marked legacy. |
| How is a unit consumed? | One request per prompt, times a per-model multiplier | Input, output and cached tokens at published per-million rates | This is the change that matters for agents. A long agent session on a big repository consumes far more than a one-line chat, and the old flat multiplier hid that. Now the length of the session is the price. |
| What does a paid plan include? | 300 a month on Pro, 1,500 on Pro+ | 1,500 credits on Pro, 7,000 on Pro+ | The numbers look bigger and are not directly comparable. 1,500 credits is $15 of model usage, not 1,500 prompts. Any article still quoting "300 premium requests" is quoting a retired allowance. |
| What does extra usage cost? | $0.04 per additional premium request | $0.01 per additional AI credit | Overage is priced identically to the included rate. There is no penalty for going over and no discount for buying ahead, which makes the plan choice a cash-flow question rather than a unit-economics one. |
| Are code completions billed? | No | No, and GitHub says so explicitly | GitHub states completions and next edit suggestions "are not billed in AI credits" and "remain unlimited for all paid Copilot plans". So the meter only ever touches chat and agents. Agent work is your entire variable cost. |
| What is the agent called? | Copilot coding agent | Copilot cloud agent, in the current docs | Both names describe the same thing: the agent that works in a GitHub Actions environment and opens a pull request. Agent mode is the separate, local one that edits files in your IDE. |
One consequence is easy to miss and worth saying plainly: under premium requests, a thirty-second question and a forty-minute agent session cost the same. Under credits they do not, and the agent session can cost a hundred times more. Every rough budget built on the old unit understates agent work. The mechanics behind the unit itself are in what GitHub AI credits actually are.
What one agent task really costs, by model
GitHub publishes per-million-token rates but no worked example, so here is one. The rates in this table are GitHub's. The token count is our assumption, stated openly so you can adjust it: one cloud agent task on a real repository, where the agent re-reads context across several turns, at 250,000 input tokens and 25,000 output tokens. The final column shows how many of those tasks a single 1,900-credit Copilot Business seat covers.
Swipe to see more
| Model | GitHub's published rate | Cost of one task | In credits | Tasks per Business seat | When to pick it |
|---|---|---|---|---|---|
| GPT-5 mini | $0.25 in / $2.00 out per 1M tokens | $0.11 | 11 credits | 168 tasks | Cheap enough that the meter stops being a planning constraint. Good for scoped, well-described tasks. |
| GPT-5.6 Terra | $2.00 in / $12.00 out per 1M tokens | $0.80 | 80 credits | 23 tasks | Mid-tier reasoning. The rate is a range in GitHub's table; the low end is used here. |
| Claude Sonnet 4.5 | $3.00 in / $15.00 out per 1M tokens | $1.13 | 113 credits | 16 tasks | The common default for agent work, and the one most teams are actually budgeting against. |
| GPT-5.3-Codex | $1.75 in / $14.00 out per 1M tokens | $0.79 | 79 credits | 24 tasks | Cheaper input than Sonnet, similar output. Agent tasks are input-heavy, so this lands lower than it looks. |
| Claude Opus 4.6 | $5.00 in / $25.00 out per 1M tokens | $1.88 | 188 credits | 10 tasks | The expensive end. Reserve it for the tasks you would otherwise assign to a senior engineer. |
That spread is the headline. One Copilot Business seat is either 168 agent tasks a month or 10, and nothing about the plan changed between those two numbers. The model picker is the single biggest cost control GitHub gives you, and it sits in a dropdown most teams never talk about in a rollout. If you want the same arithmetic across vendors rather than across models, what an AI coding agent costs per developer runs it side by side.
Agent mode and the cloud agent bill differently
Copilot has two agents and they are not metered the same way. Agent mode runs in your editor and touches one meter. The cloud agent, which GitHub used to call the coding agent, runs in a GitHub Actions environment and touches two. Six mechanics worth understanding before a rollout.
Agent mode runs locally and bills one meter
Agent mode lives inside your editor. It reads your workspace, proposes multi-file edits, runs commands you approve and iterates until the task is done, all on your machine. It is included in every paid plan and it consumes AI credits only, at the token rate of whichever model you selected in the picker. Nothing else is metered.
The cloud agent runs two meters at once
GitHub is direct about this: "Copilot cloud agent uses GitHub Actions minutes and AI credits." The agent works in a GitHub Actions-powered environment, so every session burns build minutes as well as model tokens. Inside your included Actions minutes and credits it costs nothing extra, and past them it is two separate line items on two separate meters.
Actions minutes bill the repository, not the developer
GitHub charges Actions minutes "to the repository owner, not the person who triggered the workflow runs". So a contractor with their own Copilot seat still spends your organization's Actions budget every time they delegate a task on your repository. GitHub Team accounts include 3,000 minutes a month and Enterprise Cloud 50,000, with standard Linux runners at $0.006 a minute past that.
One session, one branch, one pull request
The cloud agent "can only make changes in the repository specified when you start a task", "can only work on one branch at a time and can open exactly one pull request to address each task", and each session "has a maximum execution time of 59 minutes". That is a design decision worth knowing before you plan a large migration around it.
Credits expire every month, without exception
GitHub states that included AI credits "do not carry over between months. Unused credits are forfeited, and your allowance resets to the full monthly amount at 00:00:00 UTC on the first day of each calendar month." There is no banking a quiet month against a heavy one, which is the opposite of how several rival vendors handle it.
Organization credits pool, individual credits do not
On Business and Enterprise, "each license contributes AI credits to a shared enterprise pool" and usage beyond the pool is charged at $0.01 per credit. That is genuinely useful: two heavy agent users and eight light ones average out instead of one person hitting a wall. Individual plans have no pool and no sharing.
The Actions half is the one that surprises finance, because it lands on a different budget line than the Copilot seats and nobody connects the two until the invoice arrives. If you are comparing this against a rival that meters differently, Copilot against Claude Code sets a credit system next to a time-window one, and Cursor against Copilot does the same for per-token billing.
Which Copilot seat buys the most agent usage per dollar
Base credits match the subscription price exactly at a cent each, so the only thing that separates one seat from another on value is the flex allotment, and GitHub publishes a flex allotment for individual plans only. Line the seats up and the result is not what most procurement decks assume.
Swipe to see more
| Seat | Included a month | Value | What you are actually paying for |
|---|---|---|---|
| Copilot Pro+, $39 a month | 7,000 credits | 179 credits per dollar | 3,900 base plus a 3,100 flex allotment. The flex portion is described by GitHub as an additional amount designed to adapt as the economics of AI evolve, so it is the part that can move. |
| Copilot Enterprise, $39 a seat | 3,900 credits | 100 credits per dollar | No flex allotment is published for organization plans. You are buying the administrative surface, the shared pool, policy controls and knowledge bases, not more model usage. |
| Copilot Business, $19 a seat | 1,900 credits | 100 credits per dollar | The same cent-for-cent rate as Enterprise. Business is the cheaper administrative surface, not a worse deal on usage. |
| Copilot Max, $100 a month | 20,000 credits | 200 credits per dollar | The best included rate GitHub publishes, because the flex allotment matches the base credits one for one. It is still a single-person plan with no pooling. |
| Copilot Pro, $10 a month | 1,500 credits | 150 credits per dollar | Fifteen dollars of model usage for ten dollars, plus unlimited completions. The cheapest way to find out whether agent mode fits how you work. |
Read the first two rows together. A $39 Copilot Pro+ seat carries 7,000 credits and a $39 Copilot Enterprise seat carries 3,900, because Pro+ gets a 3,100 credit flex allotment and no organization plan does. That is 79% more included model usage for identical money. It does not mean individuals should buy Pro+ instead of Enterprise, since Enterprise sells governance rather than tokens, but it does mean nobody should price the Enterprise upgrade as a usage upgrade. We take that comparison apart properly in which $39 Copilot seat buys more AI.
The fine print buyers find out later
All nine of these are in GitHub's own documentation. Almost none of them are in the comparison articles, and two of them will change which plan you sign.
- Base credits match the subscription price exactly. At a cent a credit, $10 buys 1,000, $19 buys 1,900, $39 buys 3,900 and $100 buys 10,000. The seat fee is prepaid model spend with a product wrapped around it.
- Only individual plans carry a flex allotment. Pro adds 500, Pro+ adds 3,100 and Max adds 10,000 on top of base. GitHub publishes no flex allotment for Business or Enterprise seats.
- That produces the odd result that a $39 Pro+ seat includes 7,000 credits while a $39 Enterprise seat includes 3,900. Same price, 79% more included usage on the individual plan.
- Overage is billed at the same $0.01 a credit as the included allowance, so going over costs nothing extra per unit. Budgets are set in US dollars, not credits.
- If additional usage is capped, GitHub says "you'll need to pay off any additional usage you've already consumed in order to continue", so a hard stop mid-sprint is a payment problem, not a plan problem.
- Base credits are consumed before the flex allotment, and the same rates apply across the IDE, GitHub.com and the Copilot CLI.
- An administrator must enable the cloud agent for Business and Enterprise before any developer can start a task. Individual plans have it on by default.
- The cloud agent is available on all paid plans, including $10 Copilot Pro. There is no separate agent tier to buy.
- GitHub added AI credits consumed per user to the Copilot usage metrics API in June 2026, so per-developer attribution is available without spreadsheets.
Which Copilot plan should you buy for agent work?
One developer, trying agents properly: Copilot Pro at $10
Fifteen dollars of included model usage, unlimited completions, agent mode in the IDE and the cloud agent on GitHub. If you burn through it in a fortnight you have learned something useful about your own usage, and the upgrade path is one click.
One developer living in agent mode: Copilot Pro+ at $39
Seven thousand credits and access to the premium models agent work actually wants. This is also the best price-per-credit seat GitHub sells under $100, which is worth knowing if you are a solo consultant rather than an employee.
A team that needs policy and pooling: Copilot Business at $19 a seat
Buy this for the administrative surface and the shared credit pool, not for the usage. Ten seats give the enterprise pool 19,000 credits a month that heavy users can draw down while light users leave theirs unspent. Remember to switch the cloud agent on.
You need knowledge bases and GHEC controls: Copilot Enterprise at $39
The only reason to pay the extra $20 a seat is the enterprise feature set, because the included usage per dollar is identical to Business. Price the features, not the credits, or you will feel short-changed when the pool empties at the same rate.
You want the agent to finish the work: look past the meter
Every option above prices the model, not the outcome. If what you want is a pull request you can review rather than a faster way to write one yourself, the question to ask a vendor is what a finished task costs, not what a token costs.
Where we fit, and where Copilot beats us
We sell a competing product, so take the losses first. Copilot is inside every editor your team already uses, it has unlimited free code completions that nothing on our side replaces, and its cloud agent is wired into GitHub Issues, Actions and branch protection more tightly than any third party can manage. If your organization is standardized on GitHub Enterprise Cloud and wants one vendor, buy Copilot. We would.
What we do differently is the shape of the bill. Agentcode takes a task against a connected GitHub or GitLab repository, plans it, edits the code, runs your test suite and opens a pull request for a human to read. It never merges on its own, so your branch protection and your reviewers stay exactly as they are. It costs $29 a month, flat. There is no credit balance to watch, no model picker that changes your invoice by 17 times, and no second meter running on Actions minutes in the background.
That is a narrower product than Copilot and we are happy to say so. It is the right one if what you keep wanting is finished, reviewable work rather than a faster way to type, and if a fixed line item matters more than a bigger menu. If you are still building a shortlist, the GitHub Copilot alternatives worth considering covers the field honestly, including where we lose.
GitHub Copilot agent pricing: the questions buyers actually ask
How much is GitHub Copilot agent?
There is no separate agent price. Agent mode and the cloud agent are included in every paid Copilot plan and draw on that plan's AI credits. So the agent costs $10 a month on Pro, $39 on Pro+, $100 on Max, $19 a seat on Business or $39 a seat on Enterprise, plus any usage past the included credits at $0.01 each.
What are GitHub Copilot premium requests?
They are the retired billing unit. Until June 1, 2026 Copilot charged one premium request per prompt multiplied by a per-model rate, with 300 included on Pro and 1,500 on Pro+. GitHub replaced that with AI credits, where one credit is a cent of real token usage. The premium requests documentation is now marked legacy.
How are GitHub Copilot premium requests counted?
They are not counted any more. Copilot now meters the input, output and cached tokens each interaction consumes, converts them at the published per-million-token rate for the model you chose, and draws that amount from your AI credits at $0.01 a credit. Session length now changes the price, which it never did under premium requests.
Is Copilot agent mode free?
It is included rather than free. Every paid plan carries agent mode at no additional charge, but the tokens it consumes come out of your monthly AI credits. Copilot Free includes limited agent access. Only code completions and next edit suggestions are genuinely unmetered, and GitHub says those stay unlimited on all paid plans.
Is GitHub Copilot coding agent free?
No, though it needs no extra purchase. The cloud agent, previously called the coding agent, is available on all paid Copilot plans and consumes both AI credits and GitHub Actions minutes. Inside your included allowances there is no extra cost. Past them you pay $0.01 a credit and $0.006 a minute on standard Linux runners.
How does GitHub Copilot coding agent work?
You assign it a task on a repository and it works in a GitHub Actions environment. GitHub describes it as being able to "research a repository, create an implementation plan, and make code changes on a branch". It works on one branch, opens exactly one pull request per task, and each session runs for at most 59 minutes.
How does GitHub Copilot agent mode work?
Agent mode runs inside your editor rather than on GitHub. It reads your open workspace, plans a multi-file change, edits the files, runs terminal commands you approve and loops until the task is finished. GitHub contrasts it with the cloud agent by noting agent mode "makes autonomous edits directly in your local development environment".
What is the GitHub Copilot premium requests limit?
The limit is now a credit balance, not a request count. Copilot Pro includes 1,500 AI credits a month, Pro+ 7,000, Max 20,000, Business 1,900 per seat and Enterprise 3,900 per seat. Past that you pay $0.01 a credit, up to whatever additional usage budget your account or enterprise has set.
When do GitHub Copilot premium requests reset?
Credits reset monthly and nothing carries over. GitHub states that unused credits "are forfeited, and your allowance resets to the full monthly amount at 00:00:00 UTC on the first day of each calendar month". A quiet month cannot be banked against a heavy one, so plan agent-heavy work inside a single calendar month where you can.
Is Copilot agent mode available in Visual Studio?
Yes. Agent mode ships in Visual Studio, Visual Studio Code, JetBrains IDEs, Xcode and Eclipse, and the billing is identical everywhere: the same AI credits at the same per-model rates. GitHub applies the rates consistently across your IDE, GitHub.com and the Copilot CLI, so the editor you choose does not change your bill.
Is GitHub Copilot an agentic coding tool?
Yes, on two surfaces. Agent mode edits your local workspace autonomously, and the cloud agent works in a GitHub Actions environment and opens a pull request. Both plan, edit and iterate rather than just autocompleting. What Copilot does not do is merge for you: the pull request still waits for a human reviewer.
Which model should I use to keep the agent cheap?
Model choice moves an agent task by roughly 17 times. Using GitHub's published rates and a 250,000 input, 25,000 output token task, the same job costs about 11 credits on GPT-5 mini and about 188 on Claude Opus 4.6. Match the model to the task rather than defaulting to the largest one available.
Last updated: August 2026
Keep researching
More on Copilot: GitHub Copilot pricing, Copilot Business seats, Copilot code review and what Copilot costs a whole team.
Priced against the field: Cursor against Copilot, Codex against Copilot, Devin Desktop pricing and every AI coding tool priced side by side.
A price that does not depend on a dropdown
Connect a GitHub or GitLab repository, describe a task, and read the pull request it opens. Flat $29 a month, no credit balance, no Actions meter running behind it, and nothing merges without you.