context spent amount goes down after model switch

[m-pa]

Description

I was working for a while with gpt-5.4 pro, accumulating over $160 of spent amount. after a while i noticed the spent amount go down after each prompt.

Plugins

no plugins

OpenCode version

1.3.9

Steps to reproduce

unknown, but i noticed it after switching to gpt-5.3.

Edit: can not reproduce by switching alone. i did notice a glitch, where, after the model switch the next prompt was made with 5.4 pro selected again, which caused me to switch the model to 5.3 again, maybe two times. maybe this is related. i will post updates when encountering this again.

Screenshot and/or share link

also happens when prompts are queued and interrupted:

instead of going negative, the element stops rendering until the model comes back with a response, then appears to work normally again, starting at $0.05 iirc

Operating System

Arch Linux x86_64 6.17.8-arch1-1

Terminal

ghostty + tmux

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [TUI] $ spent in context decreases after request #6989: [TUI] $ spent in context decreases after request — describes the same behavior where the spent amount in the context block decreases after switching models or after a request, with screenshots and similar reproduction steps.

If it is a duplicate, please consider closing this issue and adding any additional details to the existing one.