Multiple models responding to the prompt in parallel.

madhatter

The assistant sends the prompt to multiple (selected by the user) models and gives the response from each, in parallel.

The user can then compare these and gather/filter the bits that are useful/interesting/common.

I've found that there are subtle differences in responses to the same prompt from the popular leading models. I've ended up having to repeat the same query in different threads and compare them. I've done this often enough that I thought it might be a good idea to have this as a feature.

Given that the usage is being counted towards billing, this should (hopefully) not be an issue as the parallel usage is also easily accounted for. In fact I'm just doing it manually in sequence now and it'd be done automatically and in parallel but the usage cost would be the same (barring the small differences owing to the responses not being deterministic).

As an aside, I suppose if we had an API for the assistant as mentioned in another thread on this forum, someone could build this for themselves (and others too, perhaps?)

What are the exact ways that you see a user using your proposed feature? Please go into as much detail as possible, and provide examples of how other browsers/apps implement this feature, if applicable. If your feature suggestion adds on to an existing feature, how would it work into it to extend its usefulness?

Use clicks button/link for response from multiple agents in parallel
From list, user selects checkboxes for preferred agents
User gets promptbox in which user enters the prompt
Responses are generated in parallel from selected agents and presented in "tabs" where each "tab" has a label which also indicates whether response is still being generated.
Each tab subsequently separates into a separate "thread" like the usual "threads" of the Assistant.

I've done this manually by copy-pasting the prompt in different browser tabs, with different models and tried to explore each "branch" further to draw nuances out of the differences. I'm asking here if it's possible to have this initial step made mostly friction-free.

foxberg34

For creative tasks, I believe querying multiple different, cheap models might give more results and still be cheaper than a single expensive model.
This would also be one of the biggest advantages that a third party provider like Kagi would have over OpenAI.

The only other provider I know that has this feature is https://openrouter.ai/chat

My idea was that you could define a "custom assistant" that includes multiple models, since it's not something you will generally need to change on every prompt.
Showing the responses side-by-side might work if you have a wide screen, otherwise top-bottom I suppose.

conradsrc

Here's another similar request, posting my comment here as well:

1000% interested in this. My reason for wanting to do is specifically so I can make my own determination as to "which model is best for this task?" and less "wanting to get a more diverse/complete answer to my query", although I could see myself using it that way occasionally. I've done this by-hand a few times and it was always to compare outputs on exactly the same prompt so I could decide which model I was going to use going forward for a given subject, e.g. programming, advice on a subject, creative tasks, etc.