ChatGPT to Route Sensitive Conversations to Reasoning Models

OpenAI is gearing up to make ChatGPT more thoughtful in tough situations. The company says it will soon start routing conversations that show signs of acute distress to its more advanced reasoning models.

Right now, ChatGPT uses different types of models. Some are lightweight and fast, while others, like GPT-5-thinking, take their time to reason through a response. A new real-time router will decide which one to use. If the system picks up signs of crisis, it’ll automatically switch to a reasoning model trained to apply safety rules more carefully.

In moments when people are stressed or struggling, a quick, surface-level answer isn’t good enough. OpenAI believes that handing those conversations to a model built for deeper reasoning can lead to safer and more supportive responses.

The company isn’t making these changes on its own. It has pulled in experts’ recommendations from its Well-Being Council as well as a global network of more than 250 physicians. These include psychiatrists, pediatricians, and other specialists whose input is shaping how ChatGPT behaves in sensitive health-related scenarios.

This update is part of a wider 120-day plan to improve the safety of ChatGPT. Some features, like parental controls, are just around the corner. But routing chats during moments of acute stress stands out as one of the biggest changes.