OpenAI released its newest artificial intelligence model, GPT-5, last week with some big new features designed to create a more seamless experience for users. This new release comes with a whole bunch of new awesome languages! You can select from “Auto,” “Fast,” and “Thinking” to tailor your interactions to match your communication style. Most users are expected to prefer the “Auto” mode, as it automatically adjusts responses to best fit the user’s needs.
Even with all these exciting features, the launch did not go as smoothly as one would hope. On its release day, GPT-5’s model router was subject to so much contention and abuse that it was completely crippled, making it extremely non-functional. Our smart model router makes real-time decisions on which AI model will best address user prompts. It has not performed well due to technical challenges and issues. OpenAI aims for GPT-5 to be a “one size fits all” AI solution, which would ideally streamline the ChatGPT experience.
OpenAI CEO Sam Altman admitted that the old model picker was very confusing. He’s seeking a better, simpler process. He stated, “Updates to ChatGPT: You can now choose between ‘Auto’, ‘Fast’, and ‘Thinking’ for GPT-5. While the majority of users will prefer to stick with Auto, that extra control should address more specific user needs. That’s why Altman stressed that “Auto” will be for the average user. That said, others might still appreciate the better traits of different models.
Paid ChatGPT Plus and Enterprise subscribers will continue to have access to legacy AI models, such as GPT-4o, GPT-4.1, and o3.GPT-5 has implemented strict usage caps. During the use of its “Thinking” mode, users are able to send up to 3,000 messages per week. For those who are looking for even greater capacity, OpenAI has released additional options under this mode.
The addition of more customizable filters is another direct result of user feedback. Altman told TechCrunch that it’s crucial to draw lessons from what’s happened in the last year. He said that in the long term, we need to head to a world where model personalities are different for every user. OpenAI is clearly cognizant of the many different needs of their user base. They still want to improve personalization features going forward, which is a good sign.
The attachment users have to specific AI models has been highlighted by recent events surrounding Anthropic’s AI model, Claude 3.5 Sonnet. When it was removed, hundreds of people participated in a symbolic wake for the model. This experience really underscored just how emotionally connected users are to their favorite AI systems.
Nick Turley, a member of OpenAI’s team at the time, looked back on the many difficulties with the rollout. And when they do fail, he underscored, it won’t be on the first try. The other thing that he’s really proud of is their speed to pivot and iterate. This feeling, in many ways, embodies OpenAI’s priority on iterating upon and responding to feedback.
