transparency label: human-only
… was released today. I have been using it to help with my writing. It is very good – markedly better than earlier versions – yet not as reliable as 4o. Commentators put this down to problems wth autoswitching, and teething problems with rolling this out at scale.
Autoswitching in ChatGPT 5 refers to the system’s ability to automatically select the most suitable reasoning mode—quick response or deeper analysis—without requiring users to manually pick a model. When you type a query, ChatGPT 5 uses an intelligent router to assess complexity and intent, deciding whether to answer fast or engage more thorough reasoning (called “GPT-5 Thinking”) for harder problems. This streamlines the experience by combining previous models into a unified, smarter, and faster AI that adapts behind the scenes for optimal results.
[source: Perplexity, citing OpenAI and others]