Regression: selected model and its options are ignored when resending edit
After editing a message to resend it, any changes made in the model picker (search, thinking depth) are discarded and the previous model is used. Historically, I could select the desired model and configure its options (for example, enable search for Gemini or increase thinking depth for o4-mini) in the input area, click “Edit” on the message, and resubmit. The new Regenerate button offers quick model switching but omits search toggles and thinking-level selectors, so anyone needing that granularity must fall back on the old edit-and-send flow—which now ignores their selections and breaks a previously reliable workflow.
Log in to comment and vote
Comments7
Coral Deer
I now need to start an entirely new chat if I don’t want the (wrong) model’s previous response in the message history and want to ask a new model the same prompt.
Scarlet Highlighter
Guys, please this is serious, it’s really annoying, when I send a message with gemini 2.0 flash and forget to select search, there is NO WAY to regenerate that message with search enabled.
Mark Florkowski
Can you confirm, is this on beta or normal t3 chat?
Purple Blueberry
This is on beta.
Mark Florkowski
I have not been able to reproduce this, can you make a screen recording showing the behavior? jam.dev is preferred, but even a plain recording would be helpful
Purple Blueberry
https://jam.dev/c/76bb041e-b14d-4fed-8623-233544eff036
Mark Florkowski
I see, if you hover over the response itself, it is using the correct model, it just does not update the input.
We’re working on making that. aglobal state to be less confusing.