I see something similar. Cursor apparently counts the token usage of BYOK, though I use BYOK of Gemini API. The usage page looks weird. The cost on the usage page seems to include both of Cursor-provided `gemini-3-flash-preview` and that of BYOK.
Hey @yu-iskw
Based on this screenshot, I would assume that some of your requests are using BYOK, and others aren’t. Is that not the case?
@Colin I use gemini-3-flash-preview only with BYOK. I don’t turn it off. However, the usage page shows the included usage was consumed.
@Colin I am doubting sub agents don’t use BYOK, even if the used model is set to inherit.
I did some experiments with my own demo repository . When I make a plan with the plan mode, Cursor used the BYOK. Meanwhile, if we go with the plan in the agent mode, Cursor used the indluded usage.
@yu-iskw Subagents is a great theory! Going to try and reproduce it and flag it for the team.
@yu-iskw Sorry for coming back to this late. I used your repo and wasn’t able to reproduce any included usage being used when subagents were configured (and BYOK was enabled). Are you still seeing the issue?
@Colin Thank you for checking it out.
I have made sure of that on the following environment. Even the agent mode can use BYOK now. I think the issue was solved today.
Version: 2.6.19
VSCode Version: 1.105.1
Commit: 224838f96445be37e3db643a163a817c15b36060
Date: 2026-03-12T04:07:27.435Z
Build Type: Stable
Release Track: Early Access
Electron: 39.4.0
Chromium: 142.0.7444.265
Node.js: 22.22.0
V8: 14.2.231.22-electron.0
OS: Darwin arm64 24.6.0
