AI Coding & Vibe Coding Published 2026-10-23

GitHub Copilot Pricing (2026): Free, Pro, Pro+ and Max, and How AI Credits Work

How we checked: plan prices, credit allowances, and per-model rates come from GitHub's official Copilot plans page, the GitHub Docs billing pages, and the GitHub Changelog, all checked on October 23, 2026 (sources at the end). We have no affiliate relationship with GitHub. Copilot's billing changed completely on June 1, 2026, so many guides you'll find still describe the old "premium requests" system.

The short answer

  • Free ($0): 2,000 code completions a month plus a small allowance for chat and agent mode. GitHub picks the model for you.
  • Pro ($10/month): unlimited code completions and 1,500 AI credits a month for chat and agents, plus the cloud agent and code review. Verified teachers and maintainers of popular open source projects may get it free.
  • Pro+ ($39/month): 7,000 AI credits a month and the option to hand tasks to third-party coding agents such as Claude and OpenAI Codex (in preview).
  • Max ($100/month): 20,000 AI credits a month and priority access to premium models. It's for people who run agents for hours every day.
  • Business ($19/seat) and Enterprise ($39/seat): credits are pooled across the whole organization, with admin controls.

The most important thing to know: since June 1, 2026, Copilot charges by tokens, not by request. One AI credit is worth one US cent, and each chat message or agent session costs whatever the model's per-token rates add up to. A quick question to a lightweight model costs a fraction of a credit. A long agent session on a frontier model can cost dozens. Code completions stay outside this system and are unlimited on every paid plan.

Copilot plans at a glance

USD, monthly prices from GitHub's plans page and the GitHub Docs on October 23, 2026.

Plan Price AI credits per month Models Best for
Free $0 A limited allowance Auto selection only Trying Copilot, light autocomplete
Student $0 (verified students) An allowance Auto selection only Students
Pro $10/mo 1,500 (1,000 base + 500 flex) A selection of models Most solo developers
Pro+ $39/mo 7,000 (3,900 base + 3,100 flex) Premium models Daily agent users
Max $100/mo 20,000 (10,000 base + 10,000 flex) Priority access to premium models Heavy, all-day agent use
Business $19/seat/mo 1,900 per user, pooled Premium models Teams and companies
Enterprise $39/seat/mo 3,900 per user, pooled Priority access to premium models Companies on GitHub Enterprise Cloud

All plans include Copilot CLI and the Copilot app. GitHub doesn't say how many credits the Free and Student plans include. It describes them only as "an allowance."

How AI credits work

Base credits and the flex allotment

Each paid individual plan has two parts:

  • Base credits equal your subscription price ($10 buys 1,000 credits on Pro). GitHub says these "never change."
  • The flex allotment is extra usage on top: 500 credits on Pro, 3,100 on Pro+, and 10,000 on Max. GitHub says it's variable and "may change over time" as model prices change.

Copilot uses your base credits first, then the flex allotment, with no setup needed. That makes the flex part a real bonus today, but don't treat it as fixed when you plan a budget. Only the base credits are guaranteed.

What uses credits (and what doesn't)

These features use AI credits: Copilot Chat, agent mode, Copilot CLI, the Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents. Code completions and next edit suggestions don't. They're unlimited on Pro, Pro+, and Max, and capped at 2,000 a month on Free.

Copilot code review is the one cost that's hard to predict. GitHub picks the model automatically and doesn't disclose it, so the cost per review varies. Each review is also billed twice: tokens in AI credits, plus the GitHub Actions minutes that run the review. Organizations can choose to bill reviews to the organization instead of to each member's credits (a setting GitHub added on October 8, 2026).

Resets, rollover, and running out

  • Included credits reset at 00:00 UTC on the 1st of every calendar month, whatever day you signed up. Unused credits don't roll over.
  • When you run out, you can upgrade, and you pay only the price difference for the rest of the month. You can also set a dollar budget for extra usage at the same rate (a $10 budget covers 1,000 credits). Or you can wait for the reset.
  • GitHub may cap extra usage on individual plans based on your usage patterns, billing history, and account verification. You also can't buy extra credits if you subscribed through the GitHub Mobile app on iOS or Android.

What a credit actually buys

Every interaction is priced per million tokens. Here are some of the rates from GitHub's models and pricing page (October 23, 2026):

Model Category Input Cached input Output
GPT-6 Luna Lightweight $0.10 $0.01 $0.50
Claude Haiku 5.5 Lightweight $0.10 $0.01 $0.50
Gemini 3.8 Flash (promo price to Dec 31, 2026) Versatile $0.75 $0.075 $3.75
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
GPT-6 Astra Powerful $10.00 $1.00 $50.00
Claude Fable 5.1 Powerful $10.00 $0.25 $50.00

Prices shown are the default tier. Some OpenAI, Anthropic, and xAI models cost more once a prompt passes a long-context threshold, and OpenAI's newer models and all Anthropic models add a cache-write charge. Check GitHub's table for the model you use.

A worked example

To see what this means, imagine one agent step that sends 40,000 input tokens (your files and conversation) and gets back 2,000 output tokens. These token counts are our own illustration, not a GitHub figure, and we've ignored caching to keep the math simple:

Model Cost of that step Credits Steps on Pro's 1,500 credits
GPT-6 Luna $0.004 + $0.001 = $0.005 0.5 about 3,000
Claude Sonnet 5.5 $0.08 + $0.02 = $0.10 10 about 150
Claude Opus 5.5 $0.16 + $0.04 = $0.20 20 about 75
GPT-6 Astra $0.40 + $0.10 = $0.50 50 about 30

Two things stand out. First, the model you pick matters far more than the plan you're on. The same work costs about 100 times more on GPT-6 Astra than on GPT-6 Luna. Second, a single agent task is rarely one step. GitHub notes that agent mode and the cloud agent "can involve multiple model calls within a single task." A task that reads a dozen files and runs tests can take many steps like this. Caching helps a lot here: context the model has already seen is billed at the cached rate, which is 10 to 40 times cheaper for the models above.

How to make your credits last

  • Use auto model selection. Paid plans get a 10% discount on model costs when Copilot picks the model in Copilot Chat, Copilot CLI, the Copilot app, or the cloud agent. Max subscribers get 30% off with auto through October 28, 2026.
  • Save the frontier models for hard problems. Use a lightweight model for quick questions, renames, and explanations. Switch to a powerful one for tricky debugging or multi-file refactors.
  • Cap agent sessions. Copilot CLI (version 1.0.66 and later) lets you set an AI credit limit per session with /limits, or with --max-ai-credits in scripts. GitHub calls it a soft cap: a response that's already running finishes first.
  • Keep your IDE and extension updated. GitHub says older versions may show wrong model prices or usage numbers (for example, VS Code needs version 1.120 or later).

Which plan should you pick?

You… Pick Why
Mostly want autocomplete, with occasional chat Free, then Pro if you hit 2,000 completions Completions are unlimited from Pro up
Code most days and use agent mode sometimes Pro ($10) 1,500 credits goes a long way on versatile and lightweight models
Run agents or the cloud agent daily on premium models Pro+ ($39) About 4.7 times Pro's credits, plus third-party agents like Claude and Codex
Run long agent sessions for hours every day Max ($100) 20,000 credits and priority access. Upgrading mid-month costs only the difference
Buy for a team Business ($19/seat) Pooled credits, admin budgets, and a central bill

If you're comparing Copilot with other AI coding tools, the pricing models differ a lot. Claude's paid plans include Claude Code, and its usage limits reset on rolling five-hour windows (see our Claude directory entry). Cursor is a separate AI code editor with its own plan structure (see the Cursor directory entry). If you'd rather describe an app than write code, see our Lovable vs Bolt vs v0 comparison. You'll find all of these in AI Coding & Vibe Coding, and the GitHub Copilot directory entry has a quick summary of every plan.

FAQ

Is GitHub Copilot free? Yes, Copilot Free costs nothing. It includes 2,000 code completions a month and a limited allowance for chat and agent mode, with the model chosen automatically. It isn't available to developers who already get Copilot through an organization or enterprise. Verified students get Copilot Student for free, and verified teachers and maintainers of popular open source projects may get Copilot Pro for free.

What happened to premium requests? They were replaced by AI credits on June 1, 2026. The old request-based billing (300 premium requests a month on Pro, 1,500 on Pro+) now applies only to Pro and Pro+ subscribers who were on an existing annual plan and stayed on legacy billing.

Can I pay for Copilot annually? The plans pages we checked list monthly prices only, and GitHub's docs mention annual billing only for legacy subscribers. Check the checkout page if you want a yearly plan.

Does Copilot train on my code? On Free, Pro, and Pro+, GitHub says it may use your prompts, suggestions, and code snippets to train its models. You can opt out at any time in your Copilot settings, and opting out doesn't affect your features. GitHub says it doesn't use Copilot Business or Enterprise data for training.

Can I cancel any time? Yes. Cancellation takes effect at the end of your current billing cycle. If your employer later assigns you a Business or Enterprise seat, your personal plan is canceled automatically, and you get a prorated refund.

Sources

Tools in this article

Get new AI tools in your inbox

A short digest of verified new tools and price changes. No spam.