[BUG] Claude Code is not local inference friendly

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?

When Claude Code is calling the model endpoint, 5 minutes timeout limit is hard-coded.
If the model is not light speed and requires more than 5 minutes thinking on a long and complex context, then Claude Code fails after 2 attempts with error message "API Error: terminated".

What Should Happen?

Claude Code should wait long enough to give a chance for the model to produce some response.

Please make this timeout value configurable with an environment variable.

API_TIMEOUT_MS variable is used for prompt processing timeout. For prompt processing, there are 10 attempts giving a chance to finish PP phase.

The main problem is that this variable maximum value is 5 minutes, any bigger value is clamped.

The thinking timeout could use the same variable, while at the same time removing the 5 minutes clamping. At this time, API_TIMEOUT_MS is not used for the A

Error Messages/Logs

API Error: terminated

Steps to Reproduce

Connect Claude Code to a thinking model with very low performance. After 2 attempts (5 minutes each) the message appears.

Claude Model

None

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.77 (Claude Code)

Platform

Other

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Xterm

Additional Information

_No response_