Skip to main content
ChatGPTreasoningUX

Why ChatGPT’s Reasoning Toggle Was So Badly Hidden

OpenAI has indeed hidden the Thinking toggle for reasoning inside ChatGPT’s attachment menu. This seems minor, but in practice it breaks familiar workflows, wastes time, and hampers AI automation because teams cannot see a critical model setting. The result is longer sessions, higher token usage, and confusion when response quality unexpectedly drops.

Technical Context

I dug in to check whether it was a perception bug or genuinely odd UX—and yes: the reasoning toggle, aka Thinking, now opens through the + button in the input field. Not in the model selector, not next to the main modes, but right where any normal person expects file attachments.

For AI integration, this is a bad signal. When a core model feature is hidden in a tools menu, users start guessing which mode is active and why answers suddenly become shorter, cheaper, or dumber.

According to the official docs, the flow goes like this: tap +, choose Thinking, and then a separate reasoning-level toggle appears within the composer—something like Light, Standard, or Extended. So the UI is scattered across two places: activation in one menu, reasoning depth in another element next to the input line.

That’s where it lost me. If a setting affects cost, latency, and answer quality, hiding it alongside the attachment flow is downright dangerous. Users won’t perceive it as model behavior, meaning they’ll keep making the same mistake over and over.

What This Means for Business and Automation

The first impact is very down-to-earth: teams waste time hunting for the mode instead of working. The second is worse: people send requests in the wrong mode, trigger extra iterations, and burn tokens where a single proper reasoning activation would have sufficed.

The third concern is architectural. If you build AI automation on top of ChatGPT and rely on a consistent reasoning level, this UX creates chaos between user expectations and the system’s actual behavior.

I see almost no upside for power users here. The losers are teams, support staff, and anyone embedding AI solution development into real processes—not just playing with the interface in the evening.

I usually catch these issues at the scenario-design stage because, in production, a tiny UI quirk quickly turns into costly confusion. If your people are already losing time over toggles like this, let’s examine your workflow: at Nahornyi AI Lab, I can build AI automation so that critical modes don’t hide from people and don’t hit your budget without a reason.

We previously broke down Claude Opus 4.6’s extended thinking, gray lines, and context cost charts. The reasoning mode this article can’t find is exactly that same feature, now inexplicably tucked inside the attachments menu.

Share this article