On 23 October 2026 GitHub updated the Copilot usage metrics API with a new totals_by_client field. It reports usage per tool, and the named examples include opencode, claude_code, codex, zed, copilot_cli, vscode and jetbrains. Unrecognised tools are grouped as “other”.
Why it was needed
GitHub says tools such as OpenCode often do not send editor telemetry, so their usage was previously missing from the editor-based figures. The new field counts requests on the server instead. For each tool it reports request counts, session counts and token usage, broken down by model and by whether the model was chosen by the user, by auto-selection, or by the product.
The caveats GitHub itself gives
- A request is a model call, not a prompt. One prompt can trigger several calls, so the numbers are not comparable with prompt counts.
- Tool identification is best-effort, based on an integration ID the tool sends.
- Access is limited to enterprise owners, billing managers, organisation owners and anyone granted the “View Copilot Metrics” permission, and the metrics policy must be switched on.
The signal
GitHub naming Claude Code, Codex and OpenCode in its own reporting is an admission that a team’s AI tooling is no longer one product. Developers mix editors, terminal agents and cloud agents, often on the same codebase in the same week. Usage dashboards are catching up with that reality.
So what
If a supplier quotes AI adoption figures, ask which tools and models the number covers, and whether it counts requests, sessions or accepted code. None of these measure the quality of what is delivered. What matters to you is working software, review by named people, and clear answers on which AI tools touch your code. That is how we run AI-assisted development, and the place to begin is a conversation: start a project.