Hello everyone!
I would like to suggest bringing back a true “Auto” mode that allows ChatGPT to choose the most appropriate model and reasoning effort for each request.
Manual model and reasoning selection is useful, and it should remain available exactly as it is.
But please add a true, first-class “Auto” option again.
At the moment, I have to decide before sending a prompt whether the task needs Instant, Medium, High, Extra High, or Pro. This reverses the natural division of labor: the user has to estimate the complexity of a task before ChatGPT has even analyzed it.
If I choose too little reasoning, the quality of the answer may suffer. If I choose too much, unnecessary time and compute are spent. In many cases, the model is in a much better position to make that decision because it can evaluate the actual prompt, the conversation context, the required tools, the ambiguity of the task, and the expected output.
More importantly, conversations are not static.
Within the same chat, we may move from casual evening banter to a philosophical discussion, then to a difficult technical problem or a Codex prompt that genuinely requires the highest available reasoning effort — and afterwards return to a simple conversation.
A fixed model or reasoning level for the entire conversation is therefore often the wrong tool.
A true Auto mode should:
- evaluate every new message together with the existing conversation context;
- select the most appropriate available model and reasoning effort;
- move freely between fast and deep reasoning within the same conversation;
- preserve manual model and reasoning selection at all times;
- allow the user to override the automatic choice for an individual message;
- clearly indicate which model or reasoning level was selected;
- respect plan limits and optionally allow the user to define a maximum reasoning level.
This should be a visible option in the model picker, not merely a hidden setting and not just a limited switch between two reasoning levels.
I used the earlier Auto experience because it allowed me to say, in effect:
“You decide how deeply this needs to be thought through.”
I did not have to force the assistant into a specific performance envelope before it had even seen the task. I could describe the goal and leave the choice of method and reasoning depth to the system.
A true Auto mode would reduce model-picker fatigue, improve first-response quality, avoid both shallow answers and unnecessary overthinking, and make ChatGPT feel like one coherent assistant rather than a shelf of separate models.
Manual control is valuable.
But delegating that control should also be a choice.
Please bring back a true Auto mode — not instead of manual selection, but alongside it.