GitHub began usage-based Copilot billing, and some developers say they used up the AI credits GitHub grants for a month in under 24 hours.
Developers burn credits through "premium requests". GitHub counts prompts to advanced models, agent tasks, edits, and some Copilot features against the allowance. During a long agent run, you can consume many requests fast. A team that treats Copilot like an open-ended coding agent can hit the cap sooner than expected.
Team admins should check:
- Plan allowance for each seat
- Models that consume premium requests
- Per-user usage in the Copilot dashboard
- Spending limits and policy controls
- IDE settings that default to costlier models
