GitHub Copilot pricing is $0 for Free, $10 a month for Pro, $39 for Pro+ and $100 for Max, with Business at $19 and Enterprise at $39 per seat a month (GitHub, 14 November 2026). Since 1 June 2026 usage is metered in GitHub AI Credits worth $0.01 each, which replaced premium requests. Code completions stay unlimited on paid plans; chat, agents and code review draw on credits.
The price points look familiar, but the meter behind them is new. Copilot used to count premium requests and multiply them by a factor for each model. It now counts tokens, prices them at each model’s rate and converts the result into credits. A plan is therefore a monthly spending allowance plus unlimited completions, and the question for a buyer is how far that allowance goes.
One distinction before the numbers. GitHub Copilot is the coding assistant sold by GitHub. Microsoft 365 Copilot, the assistant inside Word, Excel, Outlook and Teams, and Copilot Studio, Microsoft’s agent builder, are different products with their own price lists, which we cover in our guide to what Microsoft’s office assistant costs.
Everything below was taken on 14 November 2026 from GitHub’s plans page and from the Copilot documentation on plans, usage-based billing, model pricing and bring-your-own-key.
GitHub Copilot pricing plans in November 2026
| Plan (GitHub, US dollars, verified 14 November 2026) | Price per month | Who it is for | Included AI credits per month | Key limits and admin features |
|---|---|---|---|---|
| Copilot Free | $0 | Individuals trying Copilot | An allowance, amount not stated | 2,000 completions a month; automatic model selection only; no cloud agent |
| Copilot Student | $0 | Verified students | An allowance, amount not stated | Copilot in the cloud included; automatic model selection only; no third-party agents |
| Copilot Pro | $10 | Everyday individual use | 1,500 (1,000 base + 500 flex), worth $15 | Unlimited completions; a selection of models; cloud agent and code review |
| Copilot Pro+ | $39 | Individuals using premium models | 7,000 (3,900 base + 3,100 flex), worth $70 | Premium models including Opus; audit logs |
| Copilot Max | $100 | Sustained, high-volume agent work | 20,000 (10,000 base + 10,000 flex), worth $200 | Priority access to new models and features |
| Copilot Business | $19 per granted seat | Organisations | 1,900 per user, pooled | Central management, policy control, content exclusion, audit logs |
| Copilot Enterprise | $39 per granted seat | Enterprises on GitHub Enterprise Cloud | 3,900 per user, pooled | Everything in Business; priority access to premium models and new features |
| Copilot access | $0 per seat with a GitHub Enterprise Cloud licence, or $10 per user as Copilot Standalone | Eligible enterprises that prefer pure metering | None | Same features and models as Enterprise; all usage billed from first use |
The plans page shows monthly prices only. An enterprise must choose one billing model: Copilot access cannot be mixed with Business or Enterprise seats in the same enterprise.
Is GitHub Copilot free?
Copilot Free is a real plan, not a trial. It includes 2,000 completions a month, chat and agent mode within a credit allowance GitHub does not quantify, the Copilot CLI and MCP support. Models are picked by automatic selection; the plans page names Haiku 4.5 and GPT-5 mini as examples of what it reaches. It does not include Copilot in the cloud, pull request reviews or third-party agents, and its users cannot buy extra credits.
It is available only to individual developers who do not already have Copilot through an organisation. Students who verify their status get Copilot Student at no charge; GitHub’s documentation states the 2,000-completion cap for Copilot Free only and gives no completion figure for Student.
GitHub Copilot premium requests are now AI Credits
GitHub’s legacy billing article gives the date: as of 1 June 2026, request-based billing was replaced with usage-based billing. The differences:
| Before 1 June 2026 | Since 1 June 2026 | |
|---|---|---|
| Unit | Premium request | GitHub AI Credit, $0.01 each |
| What sets the cost | One request times a model multiplier | Tokens consumed (input, output, cached) at the model’s per-token price |
| Included amount | A monthly request quota | A monthly credit allowance |
| Going over | Paying per extra request | A dollar budget for additional usage, drawn down at $0.01 per credit |
What the credit system covers: Copilot Chat, Copilot CLI, the cloud agent, Copilot Spaces, Spark and third-party coding agents. What it does not cover: inline completions and next edit suggestions, which remain unlimited on paid plans and keep working even when a user has a $0 budget.
How an allowance behaves:
- Base credits and flex allotment. On individual plans the base equals the subscription price and, GitHub says, never changes. The flex allotment is extra included usage that GitHub describes as variable and designed to adapt as model economics change. Base credits are used first.
- Monthly reset. Allowances reset at 00:00:00 UTC on the first day of each calendar month, whatever your billing date. Unused credits are forfeited.
- Discounts. Paid plans get 10% off model costs when auto model selection is used in Copilot Chat or the cloud agent, and 30% off with Copilot Fusion through 30 November 2026.
Premium requests survive in one corner. Copilot Pro and Pro+ subscribers who were on an annual plan could stay on request-based billing until it ends. GitHub notes that these users do not receive new models and features, and that they drop to Copilot Free when the annual term runs out.
GitHub Copilot Pro vs Pro+ vs Max
| What changes (GitHub documentation, verified 14 November 2026) | Pro | Pro+ | Max |
|---|---|---|---|
| Monthly price | $10 | $39 | $100 |
| Base credits | 1,000 | 3,900 | 10,000 |
| Flex allotment | 500 | 3,100 | 10,000 |
| Total monthly credits | 1,500 | 7,000 | 20,000 |
| Credit value per dollar of subscription | $1.50 | About $1.79 | $2.00 |
| Claude Opus and Fable models, GPT-6.1 Sol, GPT-6 Astra | No | Yes | Yes |
| Audit logs | No | Yes | Yes |
| Priority access to new models | No | No | Yes |
| Copilot in the cloud, code review, third-party agents | Yes | Yes | Yes |
Two things separate the tiers: how many credits you get per dollar, and which models you may spend them on. Pro is limited to a selection that includes Claude Sonnet 5.5, GPT-6.1 Terra, GPT-6 Luna and Gemini 4 Hydrogen. The most capable models start at Pro+.
Upgrading mid-month costs only the price difference, and usage so far counts against the larger allowance. If you are later given a seat through Copilot Business or Enterprise, the personal plan is cancelled automatically with a pro-rated refund.
What the models cost: per-token prices and the old multipliers
Credits are spent at each model’s token price. A sample of GitHub’s table, per million tokens:
| Model (GitHub Copilot model pricing, default tier, verified 14 November 2026) | Category | Input | Cached input | Output |
|---|---|---|---|---|
| GPT-6 Luna | Lightweight | $0.10 | $0.01 | $0.50 |
| MAI-Code-1.1-Flash | Lightweight | $0.20 | $0.02 | $1.20 |
| GPT-6.1 Terra | Versatile | $1.00 | $0.05 | $5.00 |
| Claude Sonnet 5.5 | Versatile | $2.00 | $0.10 | $10.00 |
| GPT-6.1 Sol | Powerful | $2.00 | $0.10 | $10.00 |
| Claude Opus 5.5 | Powerful | $4.00 | $0.20 | $20.00 |
| Claude Fable 5.2 | Powerful | $8.00 | $0.20 | $40.00 |
| GPT-6 Astra | Powerful | $10.00 | $1.00 | $50.00 |
Several models also have a long-context tier at higher rates: GPT-6.1 Sol doubles to $4 input above 272,000 input tokens, with output at $15.
In practice the spread is the new multiplier. A million output tokens costs 50 credits on GPT-6 Luna, 1,000 on Claude Sonnet 5.5 and 5,000 on GPT-6 Astra. A Pro subscriber’s whole 1,500-credit month is $15 of model usage, which is 1.5 million Sonnet output tokens or 300,000 from Astra, before counting input.
For the record, the legacy multipliers still published for annual subscribers on request-based billing are 0.25 for MAI-Code-1.1-Flash, 0.33 for Claude Haiku 4.5, 6 for GPT-5.3-Codex, 9 for Claude Sonnet 4.6 and 27 for Claude Opus 4.8, with each Copilot code review deducting 13 premium requests. GitHub states that multipliers do not apply to usage-based billing.
Coding agent and code review usage
Agentic features are where credits go quickly. GitHub’s billing page explains why: agent mode and the cloud agent make multiple model calls within one task, and a long session across a large codebase costs far more than a chat question. The same page points to session limits as a way to cap what one agent task can spend.
Copilot code review is billed twice over. Tokens are charged in AI credits, and the infrastructure that runs the review consumes GitHub Actions minutes. The model is chosen automatically and not disclosed, so GitHub publishes estimates instead of rates:
| Review effort level (GitHub estimates, verified 14 November 2026) | Typical range per review | Approximate average |
|---|---|---|
| Lite | $0.05 to $1 | $0.25 |
| Balanced | $0.25 to $5 | $1.50 |
| Max | $1.50 to $25 | $10 |
These exclude Actions minutes, and GitHub warns that actual costs can fall outside the ranges. If the person associated with a review has no Copilot seat, the usage is billed to the enterprise or cost centre. Reviews requested by bots, and pull requests written by the cloud agent with no billable co-author, are charged to the organisation.
GitHub Copilot Business pricing vs Enterprise pricing
| (GitHub documentation, verified 14 November 2026) | Copilot Business | Copilot Enterprise |
|---|---|---|
| Price per granted seat per month | $19 | $39 |
| AI credits per user per month | 1,900 | 3,900 |
| Pooling | Across the billing entity | Across the billing entity |
| Models | Premium models | Priority access to premium models |
| New models and features | Standard access | Priority access |
| Organisation custom instructions, content exclusion, policy management, audit logs | Yes | Yes |
| Cloud agent, agent mode, code review, third-party agents | Yes | Yes |
| Platform | GitHub organisations and enterprises | Designed for GitHub Enterprise Cloud |
In GitHub’s current comparison tables the two plans list the same features. What Enterprise adds is a larger pool and priority access. Since a credit is worth one cent, $19 buys 1,900 credits and $39 buys 3,900: the seat fee and the allowance are the same number. Choosing between them is mostly a forecast of how much each developer will consume.
Pooling. An enterprise with 100 Business users gets one pool of 190,000 credits, not 100 separate buckets, so heavy users draw on what lighter users leave. Adding licences mid-cycle enlarges the pool at once.
Overage. The “AI credits paid usage” policy is on by default. When the pool is empty, usage continues at $0.01 per credit unless an administrator has switched the policy off, in which case credit-consuming features stop until the next month.
Budgets. Spending limits can be set for the enterprise, an organisation or a cost centre, and user-level budgets put a hard stop on one person’s total consumption, including what they draw from the pool.
Copilot access is the third option for eligible enterprises: no included credits, a $0 seat for users with a GitHub Enterprise Cloud licence or $10 a month as Copilot Standalone, and every credit metered from first use.
GitHub Copilot cost for 10, 50 and 200 developers
The first three rows are list prices for 12 months. The last three add an assumption, labelled as such: that each developer consumes 5,000 credits a month, or $50 of model usage. GitHub publishes no average, so replace this figure with your own usage report.
| Route (GitHub list, verified 14 November 2026) | Per developer per year | 10 developers | 50 developers | 200 developers |
|---|---|---|---|---|
| Copilot Business, seats only | 12 × $19 = $228 | $2,280 | $11,400 | $45,600 |
| Copilot Enterprise, seats only | 12 × $39 = $468 | $4,680 | $23,400 | $93,600 |
| Copilot Standalone access, seats only | 12 × $10 = $120 | $1,200 | $6,000 | $24,000 |
| Business at an assumed 5,000 credits a month | $228 + 12 × $31 = $600 | $6,000 | $30,000 | $120,000 |
| Enterprise at an assumed 5,000 credits a month | $468 + 12 × $11 = $600 | $6,000 | $30,000 | $120,000 |
| Standalone access at an assumed 5,000 credits a month | $120 + 12 × $50 = $720 | $7,200 | $36,000 | $144,000 |
The overage arithmetic: 5,000 credits less the 1,900 included on Business leaves 3,100 credits, or $31 a month. On Enterprise, 5,000 less 3,900 leaves 1,100, or $11.
What the table shows:
- Above the allowance, Business and Enterprise cost the same. Both come to $600 per developer in the example, because each seat fee is returned in credits. Enterprise is the dearer choice only when average use falls below 3,900 credits per user.
- Included pools are modest in dollar terms. Fifty Business seats share 95,000 credits a month, which is $950 of model usage.
- Metered access costs more once people use it. The Standalone seat is cheap, but it has no allowance to absorb usage.
The cost lines a Copilot seat does not show
- Paid additional usage, on by default. Without a budget, an organisation finds out about overage on the invoice.
- Premium model choice. GPT-6 Astra output costs a hundred times GPT-6 Luna output per token.
- Long-context tiers, which raise the rate once a request passes the model’s threshold.
- Code review, averaging $0.25 to $10 a review by GitHub’s estimates, plus Actions minutes.
- Seats that sit idle. An unused seat still costs $228 or $468 a year. Its credits do reach the pool, but they expire monthly.
- A flex allotment that can change. On individual plans, a third to a half of the included credits are the part GitHub reserves the right to vary.
- App and bot usage. A GitHub App installed on the organisation can make Copilot requests that the organisation pays for.
For comparison, see how Claude Code is priced across plans and API keys, what Cursor’s seats include and how Codex limits and credits work.
When your own GPU server beats Copilot on cost, and when it does not
Copilot itself now has a local route. GitHub’s bring-your-own-key documentation says a user can configure their own model provider in VS Code, JetBrains IDEs, Xcode, the Copilot CLI and the Copilot app, that this covers locally hosted as well as cloud-hosted models, and that it works without a Copilot subscription. Requests configured this way go directly to that provider and not through GitHub’s API. Organisations on Business or Enterprise can disable the option in IDEs by policy. The editor-side configuration is in our walkthrough for VS Code with a local model.
A self-hosted model is likely to be the cheaper engine when:
- Overage is routine. If the pool empties every month, moving bulk agent work to a model you host removes that share of the metered bill.
- A shared GPU server is well used. Many developers on one server spread a fixed cost.
- The hardware is already owned, or source code may not be processed outside your network.
Copilot is likely to stay cheaper when:
- Most developers mainly use completions. They are unlimited on every paid plan and cost no credits, so a $228 seat is hard to undercut.
- Usage is uneven. A pooled allowance evens out busy and quiet developers; hardware bought for the peak sits idle the rest of the time.
- The work depends on frontier models. Open-weight coding models are not equivalent to the top hosted ones, and weaker answers cost review time.
- You depend on GitHub-side features such as the cloud agent and pull request reviews, which a local model does not replace.
Hardware prices are left out on purpose. To build the comparison properly, use our on-premises TCO method and the primer on flat versus token-metered pricing. For model choice, read which local models suit coding work; for tools, the on-prem agentic coding roundup. Teams considering a full move can read what changes when you replace Copilot with a self-hosted assistant.
How VDF AI fits
VDF AI does not run, resell or host GitHub Copilot, and it does not integrate with it. It is relevant when the question moves from one developer’s subscription to an organisation’s obligations: who hosts the model, who approves models and tools, which repositories each person can reach, and what audit record is kept.
VDF Code is the governed alternative VDF AI offers for that situation. It is an on-premises coding assistant with inline completions, IDE chat, edit-mode refactors and pull request review, and it runs in your VPC, in your data centre or fully air-gapped, using the models you approve. Access is scoped by repository, and prompts and completions are logged for later review. On-premises it is licensed by seat, and the licence does not meter model tokens. The pricing page sets out the options.
Sources
GitHub’s plans page and Copilot documentation, read on 14 November 2026.
- GitHub: Copilot plans and pricing
- GitHub Docs: plans for GitHub Copilot
- GitHub Docs: usage-based billing for individuals
- GitHub Docs: usage-based billing for organisations and enterprises
- GitHub Docs: models and pricing for GitHub Copilot
- GitHub Docs: usage limits for GitHub Copilot
- GitHub Docs: what changed with Copilot billing (legacy)
- GitHub Docs: model multipliers for annual plans (legacy)
- GitHub Docs: bring your own key for GitHub Copilot
Comparing Copilot seats with an assistant you host yourself? Read about VDF Code, or contact us to plan a governed deployment for your engineers.