Disclosure: some links on this site may be affiliate links. If you buy through one, we may earn a commission at no extra cost to you. It never changes what we write or how we rank things. Details.

GitHub Copilot AI credits explained: what 1,500 credits buys (checked November 2026)

Updated November 14, 2026 · Facts come only from the vendors' official pages listed under Sources. We have not run paid accounts on these products for this page.

Most Copilot features no longer count "premium requests". They draw down a monthly allowance of GitHub AI Credits, worth $0.01 each, and how far that allowance goes depends on which model you pick and how many tokens each interaction uses. The plans page gives you the allowance; it does not tell you what the allowance buys. Below is how the arithmetic works according to GitHub's own docs, read on November 14, 2026, with worked examples.

The short answer

• 1 AI credit = $0.01. Every interaction uses input tokens, output tokens and cached tokens. Each token is priced by the model you use, and the total is converted into credits.
• Monthly allowance: Copilot Pro ($10/month) includes 1,500 credits, Pro+ ($39) 7,000 and Max ($100) 20,000. Part of each total is a "flex allotment" that GitHub says may change over time.
• Not billed in credits: code completions and next edit suggestions, which stay unlimited on paid plans.
• Reset: 00:00:00 UTC on the first day of each calendar month, whatever your billing date. Unused credits do not carry over.
• Past the allowance: upgrade (you pay only the price difference) or set a dollar budget for extra usage at the same $0.01 per credit.
• What it buys: in our examples, a chat question costs from 0.1 credit on a lightweight model to 10 credits on the most expensive one, and a 30-step agent session costs from about 5 to over 1,400 credits. The model choice moves the bill about 100-fold.

The allowance on each plan

• Base credits "match with your subscription price and they never change", in the docs' words. The flex allotment is "designed to adapt as the economics of AI evolve". Base credits are used first; the flex part applies automatically after them. If you need a figure you can rely on, plan with the base credits only.
• Copilot Free and Copilot Student include an allowance of AI credits and use models only through auto model selection. The plans page describes Free's allowance as "limited chat and agent usage" without a number. Free includes 2,000 code completions a month; Student includes unlimited completions.
• Copilot Business and Copilot Enterprise include 1,900 and 3,900 credits per user per month. These are pooled: 100 Business users share 190,000 credits, so heavy users can draw more while light users offset them. Copilot access through GitHub Enterprise Cloud or Copilot Standalone has no included allowance and is billed from the first use. Per-seat prices are on the plans page's "For businesses" tab.

How one interaction becomes credits

GitHub prices each model per 1 million tokens, with separate rates for input, cached input and output. Some models also list a cache-write rate, a cost for storing context in the cache. The cost of one interaction is roughly:

(input tokens × input rate + cached tokens × cached rate + output tokens × output rate) ÷ 1,000,000, plus any cache-write cost, then divided by $0.01 to get credits.

A selection of rates from GitHub's table (default tier, per 1 million tokens):

Notes on the rates:

• Long context: several models list a higher "Long context" rate above an input-token threshold. GPT-6 Sol, for example, goes from $2.00 to $4.00 per million input tokens above 272K, and Claude Haiku 5.5 from $0.10 to $0.50 above 100K.
• Promotional prices: GitHub's footnotes say Gemini 4 Carbon is at a promotional $2.00 input / $10.00 output until December 1, 2026, after which it becomes $4.00 / $20.00. Gemini 3.8 Flash is at promotional pricing through December 31, 2026.
• Discounts: on paid plans, auto model selection in Copilot Chat or Copilot cloud agent takes 10% off model costs, and Copilot Fusion takes 30% off through November 30, 2026.
• The full table also lists Google, Microsoft, xAI and Moonshot AI models and an image model. See GitHub's models and pricing page.

Worked examples

Two made-up but typical workloads (our assumptions, not measurements):

1. A chat question: 6,000 input tokens (your question plus the file and instructions sent as context) and 800 output tokens, nothing cached.
2. An agent session: 30 model calls, each sending 40,000 input tokens and getting 1,500 back. That is 1.2 million input tokens and 45,000 output tokens. Agent sessions resend much of the same context on every call, so we show it twice: with nothing cached, and with 90% of each call's input read from the cache.

How to read it:

• Chat is cheap; agents are not. Pro's 1,500 credits cover about 750 chat questions of this size on Claude Sonnet 5.5, or 15,000 on GPT-6 Luna. One uncached 30-step agent session on the same Sonnet-class model uses 285 credits, about a fifth of Pro's allowance.
• Caching matters. GitHub's optimization guide says cached tokens are "typically billed at 10% of the normal input price". In the example, caching cuts the agent session by roughly 70%. Every model in this table also lists a cache-write rate, which our cached column leaves out, so treat that column as a floor.
• The cache breaks easily. The same guide says switching models mid-session, changing the reasoning level or the enabled tools mid-session, or coming back after the cache expires (24 hours for OpenAI models, 1 hour for most others) makes the next request resend the whole context at the full input rate.
• Max in the same terms: Max's 20,000 credits are about 70 uncached or 250 cached Sonnet-class sessions of this size a month.

To try your own token counts, model and plan, use our GitHub Copilot AI credits calculator. It uses the same per-token rates from GitHub's table and prices the workload on every model side by side.

Copilot code review is estimated differently

For code review, GitHub picks the model automatically and does not disclose it, so you cannot use the table above. Each review is billed in AI credits and also uses GitHub Actions minutes. GitHub publishes these estimates per review, excluding Actions minutes:

At the averages, ten Balanced reviews use about 1,500 credits, Pro's whole monthly allowance, and a single Max review uses about two-thirds of it. GitHub says consumption generally rises with pull request size and repository custom instructions, and actual costs can fall outside these ranges.

When the credits run out

• Individual plans (Pro, Pro+, Max): you can upgrade, paying only the difference between plans, and your earlier usage counts inside the new plan's larger allowance. Or you can stay on your plan and set a budget for additional usage, in US dollars: "a $10 budget covers 1,000 AI credits." Or you can wait for the reset on the first of the month.
• To make sure you are never charged beyond your plan, GitHub's docs say to create or edit a "Bundled AI credits" budget, set it to $0, and enable "Stop usage when budget limit is reached."
• Additional usage may be capped: once you reach the cap, you have to pay for the additional usage already consumed before you can continue.
• Billing timing: extra usage is billed after it happens, separately from your subscription renewal, so a charge can arrive after you cancel or downgrade.
• Business and Enterprise users who exhaust the pool see a banner offering to request more budget from their administrator; what happens next depends on the organization's policy for additional usage.
• Bought Copilot in the iOS or Android app? You cannot buy extra credits through the app store. If your GitHub account has a payment method on file, extra usage is billed to it according to your budget settings.

Ways to use fewer credits

These come from GitHub's own guide to optimizing AI usage:

• Match the model to the task. Use reasoning models for architecture and hard debugging, and lighter models for refactoring, formatting and docs. GitHub suggests planning with a strong reasoning model and implementing with a cheaper one.
• Keep context lean. Start a new conversation when you switch problems; in Copilot CLI, /compact summarizes a long session and /context shows current usage. Enable only the tools you need, because large tool sets are sent as context on every request.
• Stay on one model per session so the cache keeps working.
• Set a session limit. In Copilot CLI and the Copilot SDK you can cap the credits a single session may use; the agent stops and asks before going further. It is a soft limit and does not replace your monthly budget.
• Run subagents on cheaper models. They do not inherit the main conversation, so a lighter model is often enough.

Which plan fits

These follow from the numbers above, not from hands-on use.

• Mostly code completions, occasional chat: Pro. Completions are not billed in credits, and 1,500 credits are hundreds of chat questions even on a Sonnet-class model.
• Daily agent sessions on frontier models: at the example sizes, Pro covers only a handful of uncached sessions a month. Pro+ (7,000) or Max (20,000) are sized for this, or keep Pro and set a budget for the overflow at $0.01 per credit.
• Frequent Max-level code reviews: at about 1,000 credits each on average, they will dominate any individual plan's allowance.
• Teams: Business and Enterprise pool credits across users, which smooths out heavy and light users.

If you are still choosing between Copilot and its competitors, our Cursor vs GitHub Copilot vs Windsurf comparison sets their plans side by side, including the legacy premium requests that some annual Pro and Pro+ subscribers still have.

Check the current numbers

• GitHub Copilot plans
• Usage-based billing for individuals
• Models and pricing for GitHub Copilot
• Usage-based billing for organizations and enterprises

Sources

1. GitHub Docs: Usage-based billing for individuals — https://docs.github.com/en/copilot/concepts/billing-and-usage/individuals/billing, checked November 14, 2026
2. GitHub Docs: Models and pricing for GitHub Copilot — https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing, checked November 14, 2026
3. GitHub Docs: Usage-based billing for organizations and enterprises — https://docs.github.com/en/copilot/concepts/billing-and-usage/organizations-and-enterprises/billing, checked November 14, 2026
4. GitHub Docs: Optimizing your AI usage — https://docs.github.com/en/copilot/tutorials/optimize-ai-usage, checked November 14, 2026
5. GitHub Copilot plans page — https://github.com/features/copilot/plans, checked November 14, 2026