You Can Now Control How Long GPT-5 Will ‘Think’

When Openai announced his new model, GPT-5 last week, he probably did not expect the reactions she received. Although the previous launches were enthusiastic about the supporters of AI, the most noisy feedback was very critical.

The biggest complaint was that GPT-5 deleted all the inherited models, including the favorite GPT-4O of fans, which upset users with specific workflows as well as others who had developed an emotional attachment to the models. GPT-5 also claimed to change “intelligently” between his own models according to the user prompts, but he was not initially clear with which model you really interact, frustrating users more. In short, it was a bit in disorder.

OPENAI has been in control of damage since. The company brought the GPT-4O for paid subscribers, then returned other inherited models, which should appease users who missed the specific working method of a particular model. But GPT-5 himself was also revised, as Sam Altman announced on X: if you press the Picker Model, you will notice separate models to choose, granting you greater control over the GPT-5 performance.

Here are the different models and how everyone is intended to work:

• Car: This is the default GPT-5 model, and that initially deployed last week. The idea is that you let the model decide on the duration of “thinking” according to the complexity of your prompt.

• Fast: OPENAI says that this model provides “instantaneous answers”. Although it is certainly fast, it requires a little time, but the model does not “think” as does a reasoning model.

• Mini thought: OPENAI says that this model thinks “quickly”, and that’s true. When I tested it with the next prompt – “break down a complex scientific concept such as quantum mechanics or black holes so that it is easy to understand for a general audience” – Chatgpt says it “thought for a few seconds” before generating an answer.

• Thought: This is the most “powerful” model of GPT-5, and it takes its “thinking” time in order to offer the best possible answer. When I tried this model with the invite above, he thought for 21 seconds before answering.

In addition, if you subscribe to a chatgpt plan, you can press “inheritance of the models” to choose from GPT-4O, GPT-4.1, O3 and O4-MINI.

As Sam Altman noted in his article on X, the rate limit for GPT-5 thought is 3,000 messages per week. If you touch this, you will get an additional capacity on GPT-5 Thinking Mini. Altman thinks that Auto is the model that most users will want to use, although I am not sure. I bet most users do not want to care about behaving against the rate limits, which Auto could help. But at the same time, people who use Chatgpt could like additional control on their model. If you want Chatgpt to think of a longer answer, but the model thinks that your prompt is simple, it may not think as long as you expect – and the answer you recover may not be what you are looking for.

Disclosure: The parent company of Lifehacker, Ziff Davis, filed a complaint against Openai in April, alleging that it had violated Ziff Davis Copyrights in the training and exploitation of its AI systems.