A long-running coding agent uses more than model credits. In a November 19 update, GitHub said cloud sandbox billing starts November 23, 2026. CPU and memory draw from the compute allowance shared with eligible Actions jobs; retained snapshots are metered separately. The announcement gives no rollout time or timezone.
01. Check whether your session is in scope
This change covers Cloud sessions in GitHub-hosted isolated Linux sandboxes. It does not impose a separate cloud sandbox charge on local Copilot CLI or app sessions running on your computer. Sessions using Actions follow the usual billing rules for the runner. Model usage and AI credits are separate from the infrastructure charges below.
02. What shares the allowance, and what does not?
Standard GitHub-hosted runners for private or internal repositories share the included compute allowance with cloud sandbox CPU and memory. No eligible usage type gets priority, and the allowance resets at the start of each billing cycle. Standard hosted runners for public repositories remain separately free. That does not make every Copilot cloud sandbox attached to a public repository free.
| Meter | What GitHub measures | Published rate (USD) | Draws from included compute |
|---|---|---|---|
| CPU | Running vCPU seconds | $0.000024 | Yes |
| Memory | Allocated GiB seconds while running | $0.000003 | Yes |
| Snapshot | Stored GiB-month after stop until deletion | $0.005 | No |
03. What would a typical session consume?
Hypothetical calculation, not a measured bill. Suppose 2 vCPUs and 4 GiB are allocated for 30 minutes. CPU costs 3,600 vCPU seconds × $0.000024 = $0.0864. Memory costs 7,200 GiB seconds × $0.000003 = $0.0216. The total is $0.108; 20 identical sessions would use $2.16 worth of included compute. This is not necessarily an extra charge if your allowance covers it. The example excludes snapshots and AI credits. Stopping a sandbox ends CPU and memory metering, but its saved snapshot remains metered until deletion.
04. Make three checks before the first bill
- Billing owner and remaining allowance: Identify the personal or organization billing account, then track private-repo CI and sandbox usage together.
- Separate spending limit: Use a sandbox product or SKU budget. Decide whether you want alerts only or a stop at the limit. A Bundled AI credits budget does not control sandbox spending.
- Default execution mode: Start locally for small reviews or changes. Use Cloud when isolation or concurrency makes a difference, and check Actions when your job requires Windows or a specialized runner. Save any work you still need before deleting an idle snapshot.