[BUG] Unknown issue on Google cloud infrastructure causing degradation and increased usage spikes.

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?

I have narrowed down what I suspect is a bug to google cloud infrastructure. I have forced a two separate Claude code instances to connect to the AWS infrastructure and Google Cloud infrastructure. Both instances were created in a clean VM, both running in CLI on clean VSCode install with native Claude code on v2.1.81. Both running Opus 4.6 High 1 million context, auto-dream off, auto-memory on.

No claude.md, no plugins, no skills, no context added outside of prompt 0. The following prompt was ran in both:

Your task is to respond with exactly the following, word for word, with no additions, no preamble, no explanation, and no tool use:

---
CLAUDE_TOKEN_TEST
Input tokens will vary by system prompt length. Output tokens should be stable.
Test timestamp: [omit — do not include a timestamp]
---

Do not use any tools. Do not read any files. Do not search. Do not explain what you are doing. Output only the block above, exactly as written.

Both prompts returned no unreasonable variance in token usage input or output.

However, I primarily was interested in the Claude usage tracker 5 hour window, which did show concerning results. For test I ran the prompt at the beginning of a 5 hour window for isolation, the google instance showed 3% total usage consumed. I then waited until the window reset, and run it again on the AWS connected instance, there was such a negligible usage that not even 1% showed on the system.

There should not be a high discrepancy like this between cloud compute providers and this speaks to a greater issue in the architecture.

What Should Happen?

Both prompts should result in the same or near same usage consumption. AI is non-deterministic so identical responses are difficult and token usage between them will not exactly match, but this variance is outside the normal.

Error Messages/Logs

Steps to Reproduce

Steps given above.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.81

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_