The flat-rate era for GitHub Copilot is over. On June 1, 2026, GitHub moved all Copilot plans from subscription-based premium requests to token-based AI Credits — and if you run agent mode or PR automation, your bill may have already spiked. Developers running agentic workflows are reporting cost increases of 10x to 50x. If you only use inline autocomplete, you probably won’t notice a thing. Everyone else needs to read this.
What “AI Credits” Actually Means #
Every Copilot interaction now consumes AI Credits at a rate determined by the model you’re using and the tokens in the request. One AI Credit equals one cent. Your monthly plan includes a credit allotment; burn through it and GitHub charges overages at model API rates.
Token types that count: input (your prompt, selected files, repo context), output (the model’s response), and cached tokens (context the model reuses). Token types that do not count: code completions and Next Edit Suggestions, which remain free across all plans. If Copilot chat, agent mode, PR reviews, or codebase Q&A is in your workflow, those are fully metered. GitHub’s models and pricing documentation breaks down the per-token rates for each supported model.
What Each Plan Includes Now #
| Plan | Monthly Price | Credits Included |
|---|---|---|
| Copilot Pro | $10 | $15 (1,500 credits) |
| Copilot Pro+ | $39 | $70 (7,000 credits) |
| Copilot Max | $100 | $200 (20,000 credits) |
| Copilot Business | $19/user | $19 (1,900 credits/user) |
| Copilot Enterprise | $39/user | $39 (3,900 credits/user) |
Copilot Max is a new $100/month tier launched alongside the billing change, positioned for developers running sustained agentic workflows all day. The math on Pro deserves a second look: $10/month gets you $15 in credits. A single heavy agent session with a frontier model on a large codebase can consume a significant portion of that. The full announcement is on the GitHub Blog.
Who Actually Gets Hit #
The change is a non-event for a large portion of Copilot’s user base. Developers using Copilot primarily for autocomplete will see no change — completions are free and untouched.
Everyone else is looking at a different number. Developers running agent mode for multi-file refactors, codebase-wide Q&A, PR automation, or long-horizon agentic pipelines are hitting material cost increases. TechCrunch captured the community reaction in a May 30 headline: “What a joke.” On Reddit and GitHub’s own community forums, developers shared projections showing monthly bills jumping from $29 to $750 and from $50 to $3,000. Teams running Copilot as an automated pipeline are staring at potentially unbounded costs.
There Is a Grace Period — Until September #
GitHub built in promotional credits to cushion the transition. Through September 1, 2026, Business customers receive 3,000 credits per user per month instead of the standard 1,900; Enterprise customers get 7,000 instead of 3,900. Individual Pro users also received promotional credits for June.
The practical implication: the real economic impact for most teams won’t fully materialize until September. Teams that haven’t audited their Copilot usage yet have until then to understand their baseline consumption and decide whether to adjust their plan, reduce agentic usage, or look elsewhere.
The Alternatives Now Look More Attractive #
GitHub’s timing was unfortunate for GitHub. Cognition relaunched Windsurf as Devin Desktop on June 2 — one day after Copilot’s pricing reset — with flat-rate plans starting at $20/month. Cursor remains at $20/month with a flat-fee Composer agent included. Claude Code offers predictable costs with no metering anxiety, and anecdotally appears to be drawing the loudest switchers from Copilot’s agentic user base. Cline, the VS Code extension that routes directly to provider APIs, lets developers bypass subscription overhead entirely and pay provider token rates with full transparency.
None of these tools are free. But flat-rate pricing has a clarity that metered billing fundamentally lacks — especially when the meter runs on a model with a 200K token context window.
What GitHub Got Right and Wrong #
GitHub’s rationale is sound. Agentic coding workflows genuinely consume orders of magnitude more compute than an autocomplete suggestion. Flat subscriptions that don’t scale with usage become loss leaders at scale, and sustaining them indefinitely is not a viable business model. The transition to usage-based billing was inevitable.
The execution is messier. Launching a $100/month Max plan and simultaneously telling your existing users that agentic workflows cost 10x-50x more than they expected is a communication failure, not a pricing strategy. GitHub had the data on agentic usage patterns. A cleaner migration path — with better cost projection tools and earlier transparency — would have absorbed most of the backlash.
This is the first significant crack in the “unlimited AI for a flat fee” model that has subsidized developer productivity since ChatGPT launched in 2022. The inference costs have always been real; the subscriptions were absorbing them while companies competed for market share. That phase appears to be ending.