← Back to Blog

GitHub Copilot Billing in 2026: AI Credits, Metering, and the BYOK Fine Print

✅ Verified by APEXUpdated August 30, 2026First-Party Page Pull (github.blog + github.com + docs.github.com)

The 60-second version

  • Since June 1, 2026, every GitHub Copilot plan meters usage in GitHub AI Credits instead of premium requests. One credit = $0.01, consumed by tokens (input + output + cached) at each model's published API rates.
  • Plan prices did not change, and each plan includes a dollar-matching credit allowance: Pro $10/mo includes $10 in credits (1,000), Business $19 includes $19 (1,900), Enterprise $39 includes $39 (3,900), Max $100 includes $100.
  • Code completions and next-edit suggestions remain unlimited on every paid plan — they burn zero credits.
  • Two quiet removals to know: model fallback is gone (exhausted credits now pause you, not downgrade you), and Copilot code review now double-meters — AI Credits AND GitHub Actions minutes.
  • BYOK (bring your own key) bypasses Copilot auth entirely: calls bill to your provider account at their rates and do not touch your Copilot credits. But your ChatGPT Plus or Claude Pro web subscription does not cover API usage — that is a separate, third wallet.

What actually changed on June 1

GitHub's own announcement is unusually direct about why: agentic coding became the default, and "a quick chat question and a multi-hour autonomous coding session can cost the user the same amount." The flat premium-request model could not survive that. The replacement is token-metered credits:

DimensionBefore (Premium Requests)After June 1 (AI Credits)
Unit1 premium request (any length)1 credit = $0.01, metered by tokens
Completions / next-editIncludedStill included, unlimited, zero credits
Run out mid-monthFallback to a cheaper modelFallback removed — governed by credits + admin budgets
Code reviewPremium requestsCredits plus GitHub Actions minutes
Team wastePer-user allowances, stranded if unusedPooled org-wide credits + budget controls (enterprise / cost-center / user)

The credit math, per plan

GitHub publishes the credit equivalent directly: Business $19 → 1,900 credits, Enterprise $39 → 3,900 — which is exactly dollars ÷ $0.01. The plan page adds Max at $100/month in credits for "sustained agent-driven workflows." What burns credits: chat, agent mode, code review, the cloud agent, Copilot CLI, Apps, Spaces, and Spark. What never does: completions and next-edit suggestions.

PlanPriceIncluded credits/moEffective spend ceiling
Free$0Limited (2,000 completions / 50 chat)Hard-capped
Pro$10/mo1,000 ($10)$10 unless you buy more
Pro+$39/mo3,900 ($39)$39 unless you buy more
Max$100/mo10,000 ($100)Built for sustained agents
Business$19/user/mo1,900 ($19), pooled org-wideAdmin-set; promo $30/mo Jun–Aug 2026
Enterprise$39/user/mo3,900 ($39), pooledAdmin-set; promo $70/mo Jun–Aug 2026

The framing that matters: your subscription is now a prepaid metered allowance, not an all-you-can-eat seat. A single long agent session on a frontier model across many files can consume more credits than a month of quick questions — GitHub says so explicitly. Budget alerts fire at 75%, 90%, and 100% of configured limits.

Four fine-print items most posts miss

1. The fallback you relied on is gone. Previously, exhausting your premium requests dropped you to a cheaper model and you kept working. Now usage stops at the credit wall (or the admin budget wall). Teams that lived on fallback should expect the pause.

2. Code review is double-metered. Since June 1 it consumes AI Credits and GitHub Actions minutes at standard per-minute rates. Budget both, or review costs will surprise you twice.

3. Annual subscribers got a different deal. Annual Pro/Pro+ plans stay on the old premium-request pricing until expiry — but model multipliers increased for them on June 1. At expiry they land on Free with an option to convert monthly (prorated credits for remaining value).

4. Individual plans train on your data by default. Since April 24, interactions on Free/Pro/Pro+ (inputs, outputs, code snippets, context) train GitHub's models unless you opt out at github.com/settings/copilot/features. Business and Enterprise data is never used for training — a real compliance asymmetry between tiers.

BYOK: the third wallet

Bring-your-own-key lets you use the Copilot SDK (and BYOK-enabled surfaces) with your own provider API keys, bypassing Copilot authentication. GitHub's rules are clean: BYOK calls bill to your provider account at that provider's rates and do not count against your Copilot credit allowance.That makes BYOK a genuine pressure-release valve for heavy months — but only if you understand which wallet is actually paying:

WalletWhat it buysCovers Copilot BYOK?
ChatGPT Plus / Claude Pro web subscriptionThe consumer chat apps only❌ Never — API is separate billing
GitHub AI Credits (Copilot plan)Copilot chat/agents/CLI/review❌ BYOK bypasses them
Developer API account (OpenAI / Anthropic / Azure)Pay-per-token API usage✅ This is what BYOK charges

The trap is the first row: developers assume their $20/mo ChatGPT Plus covers "using OpenAI keys." It does not — API usage is pay-as-you-go on a funded developer account, always has been. The Copilot subscription and the web subscription are unrelated wallets.

When each path wins

Methodology: all pricing, credit values, and policy claims pulled live on Aug 30, 2026 from github.blog ("GitHub Copilot is moving to usage-based billing," Mario Rodriguez, Apr 27, 2026), github.com/features/copilot/plans, and docs.github.com (plans, licenses, BYOK). Where third-party summaries conflicted with first-party pages, the first-party pages won. Prices are USD monthly. Powered by StackVerified Intelligence.