[BUG] System does not accurately report Token usage in Claude Code VS Code IDE. This is a known bug since THE BEGINNING and has never been addressed.

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

To: Anthropic Support / Claude Code Team Subject: Critical Token Counting Discrepancy in Claude Code CLI - 1 Year Ongoing Issue Body: I'm reporting a critical operational issue with Claude Code (VSCode extension) that has persisted for approximately one year and significantly impacts trust and usability.

Problem: The token counting information provided by Claude Code is consistently inconsistent across sessions.

In the current session alone:
VSCode extension status bar displays: 72% tokens consumed
Claude reported: "43% remaining" (which would be 57% consumed)
Claude has given different percentages across multiple sessions over ~1 year when asked the same question
When confronted, Claude explains it cannot separate the token counter in the IDE from the regular Claude toekn counter. It can only report the regular Claude chat token counter, but the IDE obviously has access to different numbers (that are also often wrong because of the display not updating smoothly or with timeliness.)

Impact:
Financial uncertainty: Cannot accurately track API usage costs
Planning inability: Cannot determine when context window will reset or how much work can fit in a session
Trust erosion: If the AI cannot accurately report its own fundamental operational metrics, how can I trust it with complex code generation?
Time waste: Repeated attempts to get accurate information across multiple sessions
What Claude Reports:
Claude states it has no reliable way to determine actual token usage
Claude admits discrepancy between internal counters and VSCode extension display
Claude confirms this is a known issue but has no mechanism to report bugs to development team
Claude has given different explanations across sessions for how token counting works

What I Need:
Accurate, consistent token usage reporting in the VSCode extension
Clear documentation on when/how the context window resets
Explanation for why this fundamental operational metric has been unreliable for a year
A feedback mechanism so Claude itself can report operational failures to your team

Environment:
Claude Code VSCode Extension
Windows platform
Using Claude Sonnet 4.5 model
This is a systems-level failure that undermines the entire value proposition of the tool.

What Should Happen?

When I ask Claude Code in the VS Code IDE what the token budget is at and when it resets IT SHOULD GIVE MORE THE CORRECT ANSWER OR AT LEAST NOT SUBSTITUTE THE WRONG COUNTER NUMBER AND THE WRONG WINDOW RESET CRITERIA ('just start a new conversation to reset the token' bullshit).

Error Messages/Logs

Steps to Reproduce

Go into VS Code IDE and ask the Claude Code extension how many token are left in the budget and when the counter resets.

Claude Model

Sonnet (default)

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

Claude Code VD Code IDE

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

VS Code integrated terminal

Additional Information

The fact you have not automated this system so that Claude can give you this information automatically speaks volumes on how unreliable this tool is for real world work. If it was as impressive as you sell it to be, auto reporting of problems on request of used would have worked by the second draft.

<img width="1524" height="560" alt="Image" src="https://github.com/user-attachments/assets/4be169d1-ed6c-41fd-b38c-43ef12419352" />