I build a plan with a model eg. Gemini-3-pro.
When i click build plan, the model automatically switches to “Auto”.
It happens way to often to be a mistake. this happens in the latest version.
It’s really a violation of users’ trust. If I want to use a model then use that model, give me an error but don’t bloody switch the model because you want it so.
Perhaps there should be a setting that prevents automatic model switches.
If I want to run with “Auto” I’ll chose that option myself.
No wonder the code produced was rubbish.
Steps to Reproduce
Create a plan with Gemini and then try to “Build” it. The agent starts with “Auto” model.
Expected Behavior
The agent should be locked with the user designated model. If there’s a problem i expect an error not an automatic switch to “Auto”!
Hey, thanks for the report. This is a known issue. The model can actually switch back to Auto when building a plan.
So the team can check if this is the same bug or a regression, can you share the Request ID? You can find it in the chat context menu (three dots in the top right) > Copy Request ID.
Workaround: After you pick a model and click “Build plan”, double-check the model selector before the agent starts. If it switched to Auto, change it back manually.
It’s possibly the same issue.
However the model for Build goes to “Auto” in my case. It happened in the latest version. This is super annoying. Indeed it can be fixed manually if you don’t get carried away. Clicky clicky.
Hope it gets fixed soon.
RequestID: 697302a6-e5a9-4232-9892-da5ef59e9ea4
Have the same problem. Auto keeps failing on the same problem, switched to a stronger model to fixed the problem. After awhile, the same problem comes back.
After inspecting, notices the model magically switches back to Auto and fails back at the exact problem.
This is BS and time wasting. I choose auto when I want to.
I rather get rate limited on a stronger model, than switching to a weak model I knew it wasnt able to complete my task
The original issue in this thread was fixed in v2.5.
If you’re still facing an issue on v3.0.12+ (or whatever the latest version is, if you’re in the future!), please raise a new thread and fill out a complete bug report (including version and Request ID). Any steps that help us reproduce the issue are invaluable.