How to compare two AI answers, and why they are not side by side
There is no split screen. Branching sends your question to a second model and opens the reply in its own thread. How to run the comparison properly.
You have an answer and want to put the same question to a different model.
Most people arrive here expecting a grid. Two panes, two replies, one screen, the differences picked out for you.
MultiChats does not have that view. Nothing in the app puts two answers together in one window.
What it has is branching. Branching copies your conversation into a second thread and asks your question again there, with whichever model you choose. Two answers, two threads, both saved. The comparing part is yours to do, and the rest of this page is about doing it well.
The move is called branching
Under any reply sits a row of small buttons. One of them forks the conversation.
Open it, pick a model, and three things happen at once. Every message up to that reply is copied into a new thread. Your question goes to the model you picked. You land in the new thread while the answer is still arriving.
The new chat takes the original title with "(Alternative)" added. The original is untouched, exactly as you left it. Choosing the model is the send; there is no separate confirm.
You can also branch to the same model that just answered. It sits at the top of the menu marked "(Previous)". Useful when you suspect the first reply was a bad roll rather than a wrong opinion.
One button keeps your answer, its neighbour deletes it
Beside the fork is a second button that also re-runs the question. That one replaces the reply in place. The old answer disappears the moment you tap it, the server deletes it, everything below it in the thread goes too, and nothing in the app brings any of it back.
For comparison work, that is the wrong button every single time.
Tell them apart by the icon and never by the word. The fork is a line splitting into two paths. The replacing action is a circular arrow. Those two icons are identical on web, iOS and Android.
The words are not. On the web, the fork menu is headed "Try Again". On a phone, "Try again" is the replacing action, and the fork is "Branch here". Same two words, opposite outcomes, depending on what you are holding.
If you want two panes, open a second tab
There is a menu command for half of this, and your desktop does the other half.
In the web sidebar, right click the branched chat, or open its row menu, and choose "Open in new tab". The thread opens in a tab of its own. Then snap the two windows beside each other the way your operating system already does it.
That is as close to side by side as this gets. The app hands you the second tab, and the arranging is yours.
On a phone there is no equivalent. You read one thread, then the other. Which makes the next part matter more than it looks.
Decide what you are comparing before you switch
Two threads means you are comparing from memory, and memory is generous. It keeps the tone and loses the content.
So name your target before you branch. Pick the one thing the second answer has to speak to: the figure, the recommended option, the step you doubted, the warning you suspect is missing.
Then read the second answer for that one thing first. Read it properly afterwards if you like. Going in with a target is what stops the exercise from becoming "they both sounded fine".
Retyping the question breaks the comparison
There is a temptation to open a fresh chat and paste the question in. Resist it.
A branch carries the real history: your original wording, the earlier turns, any files you attached, the corrections you made along the way. The second model starts where the first one started.
Retype it and you get a question that reads the same on top of a context that is not the same underneath. Any difference you spot could be the model, or it could be your paraphrase and the missing setup, and you will have no way to tell which.
Three comparisons worth running
A factual question. Look at the figure and at whatever date is attached to it. Two answers can be equally confident and land on different values.
A judgement call. Which option to take, whether an email reads badly, how to price something. The verdict matters less than the reasons under it. Two models often recommend the same thing for different reasons, and the reasons are the part you can check yourself.
A piece of writing. Read the opening two sentences of each and stop there. Voice announces itself immediately and then disappears into the middle of a long draft.
Every branch is a send
A branch costs what asking the question again costs, because asking the question again is what it does. It counts against your message allowance like any other send, and a free account gets 50 messages every 24 hours.
So branch selectively. The moment for it is when the answer is about to leave your hands: a number going into a document, a decision that is awkward to reverse, a reply you cannot unsend.
Questions people ask about this
Can I see two AI answers on one screen?
Not in one view. On a desktop, open the second thread in its own browser tab and place the windows side by side yourself. On mobile, you move between two chats.
Does the second model see my earlier messages?
Yes. Branching copies the conversation up to the point you branched from, so the second model gets the same history the first one had. It does not see the first model's answer, which is the point.
Where does the second answer end up?
In your chat list, under the original title with "(Alternative)" on the end. On the web a small fork icon sits next to it; hover for the parent chat's name, click to jump back. On mobile, open the chat info sheet and tap the entry under "Branched from".
How many models can I compare against?
The branch menu carries all 60 current models from 18 providers, grouped by provider, plus a few retired names marked with a warning sign. On a free account, 21 of the 60 open and the rest are marked as paid. Pricing is here.