- GitHub replaces all Copilot flat-rate plans with usage-based AI Credits on June 1.
- Business customers pay $19/month but receive only $19 in tokens after a three-month promotional buffer.
- Code completions stay free—everything else, including agentic workflows, burns credits per token.
GitHub confirmed on April 27 that every Copilot subscription—Pro, Pro+, Business, and Enterprise—moves to usage-based billing on June 1, 2026. Instead of monthly request bundles, users will receive an allotment of AI Credits debited by token consumption per model. The shift ends three years of flat-rate pricing that made Copilot the default AI coding tool for an estimated 15 million developers.
The commercial logic is straightforward. When Copilot launched in 2022, it offered single-line autocomplete. By 2026, developers are running multi-file agentic edits through Claude Opus and GPT-5.4—models that cost dramatically more per query than simple completions. GitHub was absorbing the variance under flat pricing. As reported Thurrott, GitHub’s Mario Rodriguez said the transition reflects “the incredible growth in the usage of agentic coding,” where “the compute required for those tasks is significantly higher.”
Internal documents obtained by Ed Zitron at Where’s Your Ed At revealed the promotional buffer: Copilot Business customers paying $19 per user per month will receive $30 in pooled AI credits for June, July, and August. Enterprise customers at $39 per month get $70 in credits during the same window. After August, the buffer disappears and users receive tokens worth exactly their subscription fee—$19 buys $19 in compute, $39 buys $39.
What Token Billing Means for Developer Budgets
The pricing architecture matters more than the headline. Under the old system, Copilot Pro offered 300 premium requests per month regardless of which model you selected. Under the new system, each model burns credits at its published API rate. GPT-5.4 costs $2.50 per million input tokens and $15 per million output tokens, according to GitHub’s published rate card. Claude Opus 4.7 runs at $5 per million input and $25 per million output—roughly double the GPT rate. A developer running Opus-heavy agentic workflows through large codebases could exhaust their monthly credits in days, not weeks. Neowin noted that model multipliers will also increase for users who remain on legacy annual plans—GPT-5.4 moves from a 1x to a 6x multiplier starting June 1, making the old pricing structure punitive enough to force migration.
GitHub softened one edge: code completions and Next Edit suggestions remain included in all plans without consuming AI Credits. The cost center is agentic workflows—Copilot Workspace sessions, multi-file refactors, and chat-based coding where the model generates thousands of tokens per interaction. Copilot code reviews will also consume GitHub Actions minutes alongside AI Credits, adding infrastructure cost on top of model cost.
The competitive picture shifted overnight. OpenAI’s Codex, Anthropic’s Claude Code, and Google’s Gemini Code Assist all price on consumption natively—they never promised flat-rate simplicity. GitHub’s move eliminates one of Copilot’s clearest advantages: a per-seat price that enterprise procurement teams could approve without modeling token burn. Neowin reported that the subscription model for AI coding was effectively unsustainable at $20-$30 per month, given the compute costs of modern agentic models—something Anthropic discovered separately when it imposed new rate limits on Claude Code Pro users earlier in April.
Enterprise teams face the sharpest adjustment. CFOs who approved a clean $39-per-user line item will now receive pooled credit burn charts with variance they cannot easily forecast. GitHub is offering admin controls at the enterprise, cost center, and user level to manage budgets—but building AI cost governance is a new discipline most organizations have not developed. A preview bill arrives in early May, giving teams roughly three weeks to model actual consumption before the June 1 transition.
The Register reported that April’s rate-limit tightening on Copilot Pro—along with the suspension of new Pro free trials—was a dry run for the full billing shift—the same rate-limit revolt Frontierbeat covered on April 16. GitHub suspended individual sign-ups for Pro, Pro+, and Student plans two weeks ago, citing the need to “better serve existing customers.” The sequence looks deliberate: tighten usage first, then reprice it.
GitHub will prepare a preview bill in early May so users can estimate projected costs before the June 1 transition date.
