共有:

GitHub Copilot · VS Code

Copilot spending
finally shows itself.

Ever since GitHub switched Copilot from premium-request quotas to per-token billing in June 2026, the same complaint kept coming back: "I have no idea how much I'm spending this month." VS Code now surfaces the token cost of each request right inside the editor. The meter is open.

AI Navigate Editorial2026.07.086 min read

BEFORE · INVISIBLE ? tokens · ? USD Wait for the invoice AFTER · VISIBLE 2,458 tokens · $0.031 Shown per request
01

The Blackout

June's billing change
turned the meter off

Copilot billing switched in June 2026 from a monthly premium-request quota to per-token usage that sums input and output tokens per model. Under the quota model, the admin page showed something crisp: "47 requests remaining this month." Under usage, that number is replaced by a dollar figure that moves as you type.

The problem: for weeks after the switch, neither the editor nor the web dashboard showed real-time usage. Heavy users described watching a normal-feeling day and then getting hit by the invoice. Agent mode and long prompts, in particular, can vary wildly in cost per request, and the gap between what your fingers thought they did and what the bill said was widening.


02

Where It Shows

Where and how
the number appears

You don't have to open a separate dashboard — it lands inside the editor, right next to the code you're writing.

workspace.ts — VS Code function calculate( price: number, qty: number, ) { return price * qty; } COPILOT CHAT This function needs a discount parameter… This request 2,458 tokens · $0.031 Now visible
FIG. When a Copilot Chat / Agent request completes, the token count appears in the pane right beneath it.

VS Code's Copilot now shows the input and output tokens consumed for each Chat and Agent request. It's the visibility layer that lines up with June's move to token-based billing. What used to require waiting for an invoice is now readable in place. You can react — "this prompt is too heavy, let me trim it" — without leaving the editor.

The benefit is largest when you're passing long context and when running Agent-mode loops. Prompts that dump an entire file tree, or agents that retry until they succeed, can spike into tens of thousands of tokens for a single task — and until now that only showed up as a vague "feels heavy today."

03

By The Numbers

What changes when
the meter is on

Per req
Usage shown at each response
Mid-month
No need to wait for the invoice
Agent
You feel the weight of each run

Two practical effects. First, you get a feedback loop on prompt design. "I didn't need that much context after all." "Attaching that file was wasteful — a link would do." Those judgments used to require bouncing between the web dashboard and the editor; now you can iterate on prompts while watching the counter drop.

Second, writing team guidelines gets easier. Instead of vague "don't go too crazy with Agent mode," you can say "keep an Agent task under ~10K tokens, and scope refactors up front before letting Agent run." Concrete numbers, drawn from actual measurements rather than gut feel.

04

Who It Matters To

Who benefits, who can ignore it

Real payoffBasically noise
Daily heavy users of Agent mode or long contextPeople who only trigger a few Tab completions a day
Individuals whose bill has become unpredictable since the switchUsers covered by an enterprise plan with no personal budget concern
Dev leads writing Copilot usage guidelines for a teamTeams still piloting Copilot with no settled workflow
Engineers in large monorepos who tend to attach a lot of contextSingle-file or small-project work

05

So What

Once you can see it,
how you use it changes

Taken in isolation, this update is a small quality-of-life win. What matters is that it answers a real distortion — "usage-based billing with no visible usage" — with the obvious right fix. From GitHub's side, whether users are comfortable staying on usage billing is upstream of the whole Copilot business shape. Visibility is a precondition for that trust.

The practical play for a user: for the first week, use Copilot normally and let yourself learn what your patterns cost. Once you know which actions dominate your bill, tighten prompts, tune Agent granularity, and rework how you attach context. That's how "keep the speed and quality; only cut the wasted tokens" moves from vague slogan to something you can actually do.

Source: GitHub Copilot · AI Navigate — Daily Update · 2026.07.08