ChatGPT's new Auto, Fast, Thinking modes for paid GPT-5 users
OpenAI is rolling out new operational modes for its flagship chatbot, ChatGPT, offering users the ability to select between “Auto,” “Fast,” and “Thinking” options. However, these enhancements are exclusively available to subscribers of paid tiers, including ChatGPT Plus, Pro, Teams, and Enterprise, leaving free users without access to the latest functionalities.
This update follows a period of user dissatisfaction that emerged shortly after the debut of GPT-5, OpenAI’s latest and most powerful AI model. Many paid subscribers voiced a preference for the previous GPT-4o model, with some even canceling their subscriptions in response to GPT-5’s initial performance. OpenAI CEO Sam Altman acknowledged these concerns, admitting that GPT-5’s performance was, in some instances, worse than its predecessor, despite its underlying advancements. The initial design of GPT-5 incorporated a “real-time router” intended to automatically determine the most suitable model for a given task, an approach that proved less effective in practice than anticipated. Altman had previously committed to greater transparency regarding which AI model was processing user queries.
The newly introduced modes aim to address these performance and control issues. While the “Auto” mode is designed to suffice for the majority of users, the “Fast” and “Thinking” options provide more granular control over ChatGPT’s response generation. The “Thinking” mode, in particular, is associated with an increased rate limit, allowing for up to 3,000 messages per week. This mode also boasts an expanded context limit of 196,000 tokens, enabling the AI to process and understand a significantly larger volume of information within a single conversation. Should users exhaust their weekly message quota in “Thinking” mode, the system will automatically switch to “GPT-5 Thinking mini” to continue providing service, though OpenAI notes that these rate limits may be adjusted based on overall usage patterns.
Beyond performance and control, OpenAI is also addressing user feedback regarding GPT-5’s conversational style. Earlier observations suggested GPT-5 had a somewhat “cold” personality. Altman indicated that the updated model “should feel warmer than the current personality but not as annoying (to most users) as GPT-4o,” signaling an effort to fine-tune the AI’s interaction style for a more agreeable user experience. Furthermore, in a move that directly responds to earlier user demand, paid subscribers now have the option to revert to GPT-4o if they prefer its characteristics. This flexibility, however, remains exclusive to the paid tiers, as free users are still unable to switch to legacy models like GPT-4o. These strategic adjustments highlight OpenAI’s commitment to refining its flagship AI offering, balancing cutting-edge capabilities with practical user needs and preferences.