Effective June 1, 2026, GitHub Copilot drops its flat-rate request model and moves to GitHub AI Credits — token-based billing where every chat message, agentic run, and CLI call draws from a shared credit pool. New sign-ups for Pro, Pro+, and Student plans are already paused as of April 20.
One AI credit = $0.01. The cost of using it depends on which model you're talking to and how many tokens flow through the session.
What actually changed
The unit of billing shifts from requests to tokens. Previously, Copilot plans sold "premium requests" — 300/mo on Pro, 1,000/mo on Enterprise. Now it's credits consumed by token volume × model price.
Code completions stay free. Inline completions and next edit suggestions are explicitly excluded from AI Credits and remain unlimited on all paid plans. This change is only about interactive features.
What does get billed: Copilot Chat, Copilot CLI, the cloud agent, Copilot Spaces, Spark, and third-party coding agents.
Included credits per plan:
- Copilot Business: 1,900 credits/user/month
- Copilot Enterprise: 3,900 credits/user/month
- Promotional for existing customers (June–September 2026): 3,000 / 7,000 respectively
Credits are pooled, not individual buckets. 100 Business users = 190,000 shared credits. Power users draw more; light users offset it. Overages can either be allowed (charged at per-credit rates) or hard-blocked until the next billing cycle.
Why this matters
This is what the flat-rate era ending looks like.
GitHub is exposing the underlying model cost structure directly to teams — the same economics that any raw API user has always dealt with. The implication: the cost of Copilot is now a function of how your team uses it. A quick chat query is a fraction of a credit. A long agentic session on a frontier model burns significantly more.
Teams running agents, Copilot Spaces, or heavy multi-turn workflows will hit the ceiling differently than teams using Copilot mostly for completions. That difference was always there at the model layer — it's now visible on your bill.
Budget controls exist at four levels: enterprise, org, cost-center, and individual user. Setting a user budget of $0 cuts off access entirely. Hard stops are available. This is a fundamental change in how you plan and govern AI tooling spend — and it's coming fast.
What to do
- On Business or Enterprise? Calculate your per-user credit allocation (1,900 or 3,900/mo) against your actual usage patterns. The promotional credits (3,000/7,000) give breathing room through September — use that window to baseline real consumption before the promos expire.
- Running agents or Copilot Spaces heavily? Those are the high-burn features. They behave nothing like chat; model the cost explicitly.
- On Free/Pro/Pro+? Your plan is transitioning too. GitHub is contacting affected customers — watch your inbox for migration comms.
- Engineering leaders: Get budget controls configured before June 1. Decide now: hard stop on overages, or allow additional spend with a cap?
Sources: The New Stack · GitHub Docs
✏️ Drafted with KewBot (AI), edited and approved by Drew.
Top comments (2)
Token-based pricing makes sense once AI moves from occasional help to continuous usage. The challenge now is how teams manage cost when usage starts scaling with real workflows.
Soon interviews won't ask "tell me about your experience" - they'll ask "what's your average tokens-per-thought?" 🫠