Image: GitHub GitHub Copilot's Token Billing Went Live Today. Developers Are Not Happy.
GitHub Copilot's switch from flat-rate subscriptions to GitHub AI Credits took effect June 1. The community announcement thread has nearly 1,000 downvotes, and some developers are projecting cost increases of 25x or more.
GitHub Copilot moved to usage-based billing today, June 1. The change had been announced months ago, but many developers are only now confronting what it means for their actual usage.
The short version: subscription prices stay the same, but your monthly subscription now buys you a pool of GitHub AI Credits rather than effectively unlimited access. One credit equals $0.01. Code completions and Next Edit suggestions remain unlimited and don’t touch your credit balance. Everything else does.
What changed and what didn’t
The four Copilot plans and their monthly credit allotments:
| Plan | Price | Credits included |
|---|---|---|
| Copilot Pro | $10/month | $10 in credits |
| Copilot Pro+ | $39/month | $39 in credits |
| Copilot Business | $19/user/month | $19 in credits/user |
| Copilot Enterprise | $39/user/month | $39 in credits/user |
Credits are consumed based on token usage: input tokens, output tokens, and cached tokens, at each model’s published API rate. Unused credits don’t roll over. When your pool runs out, GitHub either cuts you off or bills you for overages, depending on your plan settings.
What doesn’t change: code completions and Next Edit suggestions. GitHub kept these outside the credit system, so autocomplete stays unlimited for all paid plans.
What did change: chat queries, multi-step agent coding sessions, and Copilot code review all now draw from your credit balance. The removal of fallback model routing also went into effect today. Previously, when a premium model was unavailable, Copilot would silently downgrade you to something cheaper. That no longer happens.
The community reaction
The official announcement thread on GitHub Community has 944 thumbs down against 22 thumbs up as of this morning, with over 500 comments. By GitHub forum standards, that’s an unusually lopsided ratio.
The frustration comes down to math. A Copilot Pro user has $10 in monthly credits. A single agentic coding session on a nontrivial codebase can easily cost $30 to $40. One developer reported burning through 822 credits in a single request, wiping out 54% of their monthly allotment before lunch.
Cost projections shared in the thread ranged from alarming to eye-watering. One developer estimated their company’s monthly bill would climb from $29 to $750. Another put their number at $50 to $3,000. TechCrunch’s coverage of the community reaction ran under the headline “What a joke.”
Not everyone agreed with the outrage. Some developers pushed back in the thread, arguing the extreme cost cases stem from poor prompting habits. One comment noted that costs only spiral “if you are purely ‘vibe coding’” with bloated, iterative sessions rather than precise requests. That didn’t calm the thread much.
Why GitHub made the change
GitHub’s stated reason: agentic usage patterns have changed the economics significantly. A developer using Copilot for code completions costs GitHub a fraction of what a developer running multi-hour agentic sessions costs. Under the old flat-rate model, both paid the same.
A separate, widely-shared observation in the thread asked the obvious question: “How much money was Copilot losing” under the previous model? Reports surfaced that Copilot’s infrastructure costs were nearly doubling week-over-week through early 2026, driven by the jump in agent-mode usage since tools like Cursor and Claude Code pushed agentic workflows into the mainstream.
Business and Enterprise customers get a transition credit buffer. GitHub is providing $30 to $70 in additional promotional credits for June through August to soften the shift for organizations mid-contract.
Where developers are going
Alternatives named repeatedly in the thread: OpenAI Codex (API access with predictable per-token billing), Anthropic Claude Pro ($20/month flat, no per-session meter), DeepSeek V4 Pro, and local open-source models. The common thread is predictability. Developers who moved to agent-heavy workflows are finding that $10 or even $39 per month doesn’t go far when each session costs more than the subscription.
The irony is hard to miss. Copilot’s pricing shift is happening because agentic usage grew, and that usage grew partly because Copilot helped normalize it. The billing model that made agents feel approachable is the one that got retired.
Sources: GitHub Blog, GitHub Community Discussion, TechCrunch, GitHub Docs: Usage-based billing