GitHub announced in an August 11, 2026 changelog that the AI usage report now includes a per-model token breakdown 1. Each model line lists four token counts — input, output, cache read and cache write — next to the AI credits that model consumed 1. GitHub says the earlier version of the report gave the credit totals without the underlying token detail, which made charges hard to explain or to bring down 1.
The benefit GitHub claims for the change is that a bill can now be traced back to its parts — how much of a given model’s cost came from fresh input, from generated output, and from cached context — and that the reasoning can be handed to whoever is asking about it 1. Who sees the breakdown depends on the plan: on Copilot Business and Copilot Enterprise it is limited to administrators, while on the individual plan every user has it 1. Either way, the file comes from the AI usage page under billing settings 1. The change is to what the report displays, and the changelog describes no change to how usage is billed.
Sources
- Per-model token breakdown in the usage report - GitHub Changelog (August 11, 2026)