ChatGPT spontaneously resets reasoning mode from High to Medium

ChatGPT users are experiencing a persistent technical issue where the platform's reasoning mode (thinking) spontaneously resets from the user-selected High to Medium. This unexpected behavior affects the web interface, as well as official applications across Windows, macOS, Android, and iOS, causing frustration among premium users who rely on consistent model behavior for complex tasks.

• The issue occurs during new chats, message submissions, and even within ongoing conversations.
• Systems are reverting from GPT-5.6 Sol High to GPT-6 Medium despite manual overrides.
• Complaints have been surfacing on official forums consistently since early October 2026.
• OpenAI has yet to provide an official explanation or fix for this system-wide bug.

The Evolution of Reasoning Modes In June 2026, OpenAI streamlined its reasoning interface: 'Thinking Standard' became 'Medium', and 'Thinking Extended' was rebranded as 'High', while 'Thinking Light' was deprecated. While intended to simplify user experience, this transition has been marred by technical instabilities that seem to have compromised the model's ability to retain user-specified settings.

Impact on User Workflow Reports from the OpenAI community forum indicate that the issue is not limited to specific devices. Even after manually selecting the High reasoning tier, the system consistently forces a revert to Medium. This is particularly problematic for high-stakes analytical tasks where the depth of 'thinking' directly correlates to the quality, logic, and accuracy of the AI's output.

What This Means for Users For professionals and power users, the inability to control the AI's reasoning depth creates significant unpredictability in workflows. When the system ignores user input and defaults to a lower reasoning tier, it undermines the tool's utility. Until a patch is released, users are forced to manually verify their settings frequently to ensure they are getting the performance they are paying for.