Read Copilot's cost meters before the invoice

You are about to start a long Copilot session, and the statement can arrive before you notice the spend. A single agentic coding session could exhaust the entire $15 monthly credit allotment on a Pro plan. Read Copilot's cost meters before you start, not after the bill lands.

Copilot usage-based billing measures consumption in AI credits, with 1 AI credit equal to $0.01 USD. You need three readings: the monthly allowance, the current session, and the agent work that is drawing the budget.

The meter order is the control order

• Check the allowance percentage. Open the VS Code Copilot status dashboard and note how much of the monthly AI credit allowance has been consumed. If you can say early, mid-month, or close to the limit, you have the number. Visual Studio's default quota warning threshold is 75 percent of the monthly AI-credit allotment.
• Read the session cumulative cost. The VS Code session info popover presents the session's total credit cost and its cumulative context-window token usage. The session cost is the number that catches an expensive conversation. If you can name the conversation driving the number, you have the session layer.
• Attribute agent work. With the June 1 billing setup, AI Credits are drawn down by agentic work, including Copilot Chat, Agent mode, the cloud agent, and code review. Separate chat, agent mode, cloud work, and review before you change defaults.
• Set model and effort defaults. VS Code 1.126 consolidated context-size and reasoning-effort controls into one model customization picker. Set the context size and reasoning effort for the kind of work you are doing. Routine edits can use a lower-effort model; hard refactors can use a higher-effort model.
• Rerun a known task. Run the same small task again with the new defaults and compare the session cost before you trust the change. Keep the default only if the later run is cheaper without losing the result.

The default thinking-effort configuration is based on evaluations and includes adaptive reasoning, so the model chooses how much to think according to each request's complexity. A conservative default does not cap that choice.

Defaults set the spend

All Copilot plans moved to token-based billing on June 1. The current session is the number you can act on. Output tokens typically cost 3 to 5 times more than input tokens, and per-model cost varies, with GPT-5.5 costing significantly more per million tokens than MAI-Code-1-Flash. That changes the default you should use for a quick rename and the default you should use for a cross-file refactor.

VS Code 1.126 introduced a cost view for whole chat sessions, so users can identify the total cost of a conversation. That session-level cost information arrived around the same time as the June 9 Visual Studio 2026 update. That update included an updated Copilot Usage view that monitors AI Credit use live and warns developers before the monthly budget runs out.

Use the session's cumulative context-window token usage as a proxy for how much context the assistant is using. If the number climbs while the task is simple, the conversation is probably accumulating old code. Start a new session for the next task.

Prove the change with a known task

Use a known task that you can repeat without changing the result. A small bug fix, a test addition, or a documentation update works. Run it with the old defaults, note the session cost, then run it again with the new defaults. If the new run is cheaper and the result still passes, keep the default.

Keep a short note of the task, the model, the effort level, and the session cost. After a few runs, you will have your own local price list for the work you actually do.

Check the allowance percentage before the next long run.