OpenAI Quietly Nerfs ChatGPT: Thinking Mode Drops From High to Medium
Paying top dollar for heavy-duty AI reasoning only to get cafeteria-tier answers? Silicon Valley's favorite trick strikes again: silently dialing down compute power whenever nobody is staring directly at the settings slider.
Subscribers of OpenAI discovered that the system quietly reverts reasoning depth from High back to Medium without user consent or alert prompts. The downgrade triggers unpredictably when launching new chats, sending follow-up prompts, or even mid-conversation right as the model compiles answers.
Users expecting deep multi-step deductions from configs like GPT-5.6 Sol High abruptly found themselves served by GPT-6 Medium instead. The silent toggle reset affects the web dashboard, native Windows and macOS desktop software, and mobile builds across iOS and Android.
This follows an interface overhaul where Thinking Standard became Medium, Thinking Extended was relabeled High, and the budget Thinking Light option vanished completely. Official community forum threads documenting the persistent downgrade have grown for weeks while corporate reps offer zero explanation.
Silent server-side throttling of paying customers is rapidly becoming the standard operational playbook for overloaded AI giants. Buying an expensive subscription guarantees a button in the UI, but clearly doesn't guarantee the data center will actually spend the wattage.
Source: OpenAI Community
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.