GitHub Copilot's new pricing model takes effect; many users report sticker shock, with some saying a few hours of AI usage eats big chunks of their monthly caps
In April, GitHub announced that it was moving subscribers from request-based billing to a usage-based model for its AI-powered Copilot service.
Context & Ripple Effects
Copilot began as a flat-priced coding assistant after its 2022 launch, then added “premium requests” in 2025 to limit access to models beyond its base offering. GitHub’s June transition completes a broader shift from request limits to monthly AI Credits tied to usage.
Related coverage also tied the planned billing change to rising costs to operate Copilot. Reports that users are rapidly reaching their caps make the practical consequences of that cost allocation visible immediately after the model took effect.
First-order effects
- Copilot subscribers now have to manage monthly AI Credit consumption rather than rely on the prior request-based allowance; heavy users face earlier caps or additional usage costs.
- GitHub gains a pricing mechanism that more directly links Copilot revenue to the intensity and model mix of customer usage.
Second-order effects
- Engineering teams that had treated Copilot as a predictable per-seat software expense may impose usage controls, steer developers toward lower-cost modes, or reconsider which workers receive paid access.
- The change raises the importance of transparent usage metering and model-specific pricing across coding-assistant providers, because customers can now compare not just seat prices but the usable capacity those prices buy.
Third-order effects
- If usage-priced AI coding tools become standard, developer-AI procurement could shift from broadly deployed SaaS subscriptions toward actively managed compute budgets, with adoption shaped by measurable task value rather than simple seat counts.
- The backlash illustrates a durable tension in generative-AI software: vendors need to recover variable inference costs, while customers expect predictable pricing. Whether credit systems persist will depend on whether they become legible enough for teams to budget and govern.
The trend: AI software is moving from flat per-user subscriptions toward pricing that passes more of the underlying model-compute variability to customers.