Technical Context
I dug into the complaints about the mobile ChatGPT and quickly realized: people aren’t exaggerating. The explicit model selection has indeed been replaced with simplified modes—Instant, Thinking, and Pro—and the app pushes users toward Instant by default.
To the average person, it looks like, “Well, it’s ChatGPT, it just answers.” But in reality, it’s no longer a transparent model choice; it’s an auto-router that decides how much to think, whether to search, and what stack to spin up under the hood. For AI integration, this is a bad sign: when a system hides an important toggle, people end up blaming the results, not the interface.
I also checked how to get to a proper model selector. Currently, you either go through Configure → Model, or dig into settings to enable additional or legacy models. On iPhone, some options have migrated deep into menus, and on Android, users are reporting that the selector has been “dumbed down” almost beyond recognition.
And here’s my main concern—not about UX, but about quality. Instant mode really does often respond too fast and without proper search, even when search is clearly needed. If someone doesn’t know they can switch to Thinking, they draw a simple conclusion: ChatGPT has gotten worse.
This is especially noticeable among non-technical users. They don’t think in terms of GPT-4o, o3, or routing. They see a single window and judge the entire product by its default behavior. If the default is weak, the reputation takes a full hit.
What This Changes for Business and Automation
The first consequence is simple: the gap between “AI sort of works” and “AI actually solves the problem” widens. If an employee or customer is stuck on Instant, they could be getting fast but superficial answers without even realizing the issue is the mode, not the model itself.
Second: AI automation becomes harder in products where predictability matters. If you build workflows around ChatGPT without controlling model selection, the quality of search, summarization, and reasoning starts to fluctuate.
Who wins? Only the onboarding flow for mass users. Who loses? Everyone who needs stability: support teams, sales, research tasks, internal assistants.
I’ve long followed a simple rule: if the model affects the outcome, don’t hide it from the process. At Nahornyi AI Lab, we address these issues at the architecture level, not by hoping the interface does magic. If your answer quality is slipping or users don’t trust your assistant, let’s take a holistic look at your scenario and build an AI solution that leads to the right results—not a random default.