GitHub Copilot Shows Per-Model Token Usage in VS Code Chat
GitHub CopilotGitHub Copilot's chat in VS Code now shows a per-model token breakdown for every chat turn, letting developers hover over a response's footer to see input, cached input, and output token usage for whichever model answered that turn. The feature gives developers turn-by-turn visibility into which requests and models are consuming the most tokens, complementing GitHub's existing session-level and downloadable usage reports. It shipped as part of the VS Code 1.135 release alongside other August 2026 Copilot updates like Rubber Duck review and side-by-side agent chats.
Key Takeaways
- GitHub Copilot's VS Code chat now shows a per-model token breakdown (input, cached input, output) for every individual chat turn, viewable by hovering over the response footer.
- The feature gives developers turn-by-turn cost visibility, useful when switching between a fast model and a more capable one within the same conversation.
- It complements, rather than duplicates, GitHub's existing downloadable AI usage report and the session-level credit indicator already shown in Copilot's web chat.
- InfoWorld confirmed the feature ships specifically in VS Code 1.135, released in late August 2026.
- The update landed alongside other high-profile VS Code 1.135 features, including the Rubber Duck second-opinion reviewer and the ability to arrange chats side by side.
- The change reflects GitHub's ongoing push toward granular cost transparency as Copilot usage is increasingly metered through AI credits tied to token consumption.
Sources & Mentions
1 external resource covering this update
Per-Turn, Per-Model Token Visibility
GitHub Copilot's chat in VS Code now breaks down token usage by model for each individual chat turn. Hovering over the response footer in the Agents window reveals the input tokens, cached input tokens, and output tokens that specific model consumed to produce that response. For developers juggling multiple models in a single conversation, such as switching from a fast model for routine edits to a more capable one for a tricky bug, the feature makes it possible to see exactly which turn and which model drove up token consumption, rather than only seeing an aggregate total for the whole session.
Filling a Gap in Copilot's Usage Reporting
The addition builds on other transparency work GitHub Copilot has shipped throughout 2026 around AI credit and token spend. GitHub already offers a downloadable AI usage report with a per-model token breakdown for administrators and individual subscribers, and a session-level credit view in the web chat overlay at github.com/copilot. What was missing was a lightweight, in-the-moment way to see cost while actually working in the editor. This hover-based footer breakdown fills that gap directly inside VS Code's chat interface, without requiring a developer to leave the editor or open a separate report.
Part of a Broader August Release
This token visibility update shipped inside VS Code 1.135, the same release that introduced the experimental Rubber Duck second-opinion reviewer, external agent session continuation, and the ability to arrange multiple agent chats side by side. Together with those changes, GitHub Copilot's August 2026 VS Code releases focused heavily on giving developers more visibility and control over long-running, multi-model agent sessions, of which understanding token cost per model is a core piece.