In old Assistant the user could kick off a prompt on a selected model, and then as soon as the answer/processing has started go back to the "edit" mode of the prompt, switch to a different model and process the same prompt in parallel with a different LLM.
I used this feature heavily to get the take of different models on certain questions where their performance varies a lot.
Currently I have to let the first model finish to answer, before I can start to use a different model, which is not time efficient.
Compare OpenRouter UI, where you pre-select multiple models and the prompt runs instantly against those in parallel.
I doubt there is a technical limitation for this current behavior (or that can't be worked around reasonably easily), so classifying as a bug. Definitely hampers my productivity and had been a major driver for Assistant use for me.
Once a prompt is kicked off and being processed by the selected model, I want to go back and use additional models on the same prompt. I do not want to wait until each processing/answer finishes.
If it is difficult to manage the prompt state with the old behavior now, pre-selecting different models the prompt should run against in parallel could be an alternative way to handle this.
