Copilot makes reasoning a per-task cost dial

GitHub Copilot cloud agent now exposes a reasoning-level control on five paid plan families, turning inference depth into a per-task cost choice. GitHub’s product announcement says higher levels can improve complex answers but consume more tokens and therefore more credits, and names Pro, Pro+, Business, Enterprise, and Max as supported plans. For teams already measuring Copilot adoption with active-user days rather than seat counts, the next useful metric is accepted work per credit at each reasoning level—not whether engineers habitually select the maximum.