August 13, 2025 | OpenAI is scaling back its enterprising plan to have GPT-5 automatically select the best model for user queries. The company is now offering more granular control for paid Plus subscribers, a move that could satisfy a broader range of users.
For those who prefer a hands-off approach, the system defaults to automatic model selection. Though, users seeking more control can now choose from three distinct modes: Fast, Thinking Mini, and Thinking.
- Paid ChatGPT users can now select specific AI models rather than relying solely on automatic routing.
- The “Thinking” mode features a large context window of 196,000 tokens, while “Fast” is limited to 32,000 tokens.
- Access to older models, excluding GPT 4.5, has been restored for Plus subscribers.
- Free ChatGPT users face stricter message limits and have limited access to advanced modes.
The “Thinking” mode comes with a weekly limit of 3,000 messages. Once this threshold is reached, users will transition to the “Thinking Mini” mode, though the exact number of queries allowed in this mode remains unspecified.
The “Thinking” mode boasts an impressive context window of 196,000 tokens, while the “Fast” mode is limited to 32,000 tokens. This preference for manual control is especially relevant for Custom GPTs, where users pre-define specific requirements. The introduction of GPT-5 disrupted this, initially limiting choices to a basic model and the GPT-5 thinking mode, omitting the mini or older generation options. The automatic routing isn’t always aligned with user expectations.
Selection of models in ChatGPT Plus after the transition to GPT-5
Message limits have also been adjusted. Previously, ChatGPT Plus users had 80 GPT-4O messages per three hours and 100 messages for the O3 reasoning model. GPT-5 temporarily increased these to 160 and 200, respectively, but these higher limits are expected to revert to original values.This change could lead to users exhausting their message limits in higher-tier models when they might have previously used the unlimited mini models. While the mini model is not currently excluded, it will now be automatically engaged after the primary model’s quota is depleted.
Free Version: One Step Forward, One Step Back
Stricter restrictions have also been implemented for free ChatGPT accounts. Previously, these users received 10 messages every 3 hours, after which GPT-4O would switch to the faster, less capable GPT-4O mini. Now, the limit is 10 messages every 5 hours. On the positive side, free users can now access the GPT-5 Thinking mode with one message daily and gain access to the O3 reasoning model, which was previously unavailable.
The company is actively listening to user feedback. The chief executive confirmed that ChatGPT Plus users will regain access to the older GPT-4O. They will continue to monitor usage patterns to determine which models are essential to maintain.
ChatGPT is also improving its indication of which model is currently active, with a more prominent button to manually engage the reasoning mode. The company is also considering offering a limited number of messages from the advanced GPT-5 Pro model to Plus subscribers.
