VS Code 1.126 Lets You See What a Copilot Session Costs

VS Code 1.126 shipped on June 24, and the headlining addition is direct: you can now see what a Copilot session costs, not just individual turns. Given that GitHub Copilot moved to usage-based AI Credits billing on June 1 and developers have been comparing notes on unexpected charges ever since, this was an obvious gap to fill.

Session Cost Display

The session cost view shows total credit consumption across the entire chat session, not per-message. This makes it easier to spot which conversations are expensive — a long agentic session that spawned multiple tool calls looks very different from a few quick questions, and now you can see that at a glance rather than doing math in the billing dashboard.

The previous version (1.125) added a spend meter that tracks running totals, but it showed per-turn cost. The 1.126 upgrade aggregates that into a session view, which is more useful when a session involves dozens of turns of back-and-forth.

Multiple Chats Per Session

Agent host Copilot sessions can now run several chats simultaneously. Each chat has its own conversation history but they share the same session and working context, so you can keep parallel conversations going in the same workspace.

The practical use case here is running different tasks in parallel without blowing through agent host session limits. One chat handles a bug fix, another writes tests, and both are referencing the same codebase without you needing to switch contexts manually.

Agentic Code Feedback

Comments on AI-generated code are now stored server-side rather than just in the editor. This gives agents access to feedback through a set of tools: listComments to read existing comments, resolveComments to mark them done, and addComment to create new ones.

The practical upshot is that when you leave a comment on agent-generated code — “this function doesn’t handle the null case” — a follow-up agent session can actually read and act on that comment without you having to retype it. Feedback becomes part of the task context rather than just UI annotations.

Model Customization

The context size and reasoning effort controls have been merged into a single model customization picker. Previously, these were configured separately, which required navigating two different settings dialogs. The combined picker simplifies setup, especially for developers using models that support both extended context and configurable thinking.

Restricted Mode by Default

New folders now open in Restricted Mode, a sandbox state where extensions can’t run arbitrary code and automation doesn’t trigger automatically. This was an existing VS Code feature, but it wasn’t the default for new folders — developers browsing unfamiliar repos could find themselves running extension code they hadn’t explicitly trusted.

The update also removes the “Trust Parent” button from the Workspace Trust editor. That button was a common source of accidental over-trusting, where developers would click it intending to trust a specific project and end up granting trust to an entire parent directory containing other projects.

VS Code Update

VS Code 1.126 is available now through the standard update channel. The full release notes are at code.visualstudio.com/updates/v1_126.

Sources: VS Code 1.126 Release Notes, Visual Studio Magazine