Didn't like the answer? Try again.

Things go wrong. A provider hiccups. The network drops the stream. The reply comes back empty, half-finished, or simply not what you asked for. And sometimes nothing failed at all. You just do not like the answer.

On the latest reply in a chat, there are three ways to do something about that: Retry request, Try another model, and, when Auto picked the model, Why this model?

Retry request

Under the latest assistant message there is a replay icon. Hover it, or use a screen reader, and the name is Retry request. On most failed replies the same words show up as a button in the error itself.

Press it and Ethyx sends that request again and replaces the reply, as if the last attempt had not happened. Your question stays. The answer that missed is gone.

I added this because if something can go wrong, it eventually will. We have already seen downstream errors and network glitches. Starting a new chat every time that happens is a lot of work for a problem that is often temporary. Retry is the smaller move: same question, fresh attempt.

There is a second reason, and it has nothing to do with failures. When AI makes things up walks through it. Ask a calculator for 2 + 2 and you get 4 every time. Ask an AI the same question twice and you often get two different answers. The randomness is built in on purpose, so the writing stays natural. Retry uses that. Even when the first reply arrived fine, trying again can produce a different one. For a draft or a brainstorm, that is useful. For a fact, a reply that changes when you ask again is still a warning — the same one that post already gives.

Retry is only on the latest turn. Once you have sent another message, the earlier reply stays where it is.

The button also stays hidden when repeating the same request cannot help: you are out of credits, the account does not have AI access, or this server cannot read the conversation history. A second attempt would fail the same way. The error already points at the way forward — usage, access, or a new chat.

Try another model

Try another model is the same replacement, with a choice of model in the middle. On that latest reply, open More message actions (the … menu beside copy and retry), pick a model, and Ethyx runs the request again on the one you picked. That model also stays selected, so the next message in the conversation continues with it.

I use this all the time. Auto's pick is usually a good one. Sometimes it is the one I did not want. Sometimes I tapped the wrong model. Sometimes I read the answer and can see the model was a poor fit for the question. Those are ordinary reasons. The control is there so you can find out without copying the question into a new chat.

So many models is the longer version of why a different model can matter. Each one is a different crowd of guessers, with different strengths and a different manner. Switching is how you ask the other crowd.

This one is for ordinary text chat. Compare mode and a workflow turn keep their own controls.

Why this model?

When Auto chose the model, a finished reply is marked · Auto. Open that and the dialog is titled Why this model?

This is our first attempt at telling you why, and the feedback is basic on purpose. The dialog names the lane — Writing, Coding, Research, and the rest — next to the model that ran, then gives a short reason in plain language. The wording matched a kind of task Ethyx recognizes. The conversation was already on this model and the task did not clearly change. The turn includes an image or a document. Expert search, or a long document, called for something stronger than a short reply would. There are a handful of notes like that. They describe the pick.

Under the reasons: Was this a good fit? A thumb up or a thumb down. One thumb at a time, and pressing the selected one again clears it. A thumbs-down can add what was off: Wrong answer, Wrong tone, Too slow, or Too costly.

The same dialog lists other models that could have taken the question, under Try a different model. On the latest reply those buttons say Try again with that model, which is the same replacement as the menu above. On an older reply, where retry is no longer available, the button says Use that model next, so the following message starts there. A thumbs-down also asks Want to try a different model? when those alternatives are on the list.

You may also see a credit comparison: what this reply used, what the same tokens would have cost on a flagship model, and the difference. That figure is an estimate against list prices. The dialog says it is an estimate, and that it is not a billing figure.

Here is the limit, and I want it in the open. Ethyx stores the thumb so we can study how Auto is doing. We do not train on that data. In time those signals could help us improve how Auto routes. Right now they are analytical. A thumb does not change the next pick.

The first reply does not stay

Retry and Try another model overwrite the latest answer. If you want several models answering at once, and then a winner, that is compare. If you want a second thread that keeps both histories, that is fork.