Sam Altman Says 'GPT-5 Current Modifications Are' - Connect2ConnectOnline August 17, 2025

OpenAI announced that it’s making adjustments to GPT-5 to ensure its AI-generated replies feel friendlier and more relatable.

The updates after the launch come from user feedback indicating that GPT-5’s replies seemed a bit too formal. The adjustments are minor, but ChatGPT is designed to feel more friendly now. You’ll see little, authentic phrases like “Good question” or “Great start”, which aren’t just compliments.

According to internal tests, “there’s been no increase in flattery compared to the earlier GPT-5 personality”, OpenAI mentioned in a post on X on Friday, August 15.

The changes in behavior for GPT-5 are set to be released next week.

This is just one of several updates that the Microsoft supported AI startup has revealed since the model’s debut on August 7.

Many users are feeling let down by the lackluster launch of GPT-5, especially after all the excitement surrounding the company’s 2023 release of GPT-4.

GPT-5 faced a number of delays because of safety testing and computing constraints. When it was finally released for free in ChatGPT this month, users noticed that the improvements they anticipated felt minimal, with the primary enhancements of GPT-5 being in cost and speed.

OpenAI’s CEO Sam Altman allegedly mentioned to reporters at a dinner last week, “I think we totally messed up some aspects of the rollout”.

Regarding the behavioral issues of GPT-5, Nick Turley, the product head of ChatGPT, mentioned, “GPT-5 was very straightforward. I appreciate that. I use the robot personality – I’m German, you know, whatever. But a lot of people don’t, and they really enjoy that ChatGPT actually checks in with them”.

Here’s a quick rundown of all the changes and enhancements that OpenAI has announced since the release of GPT-5.

New GPT-5 Modes

GPT-5 has three modes – Auto, Fast, and Thinking.

Fast mode provides quicker responses from GPT-5, while Thinking mode allows the model to take its time for more in-depth answers. Auto mode automatically switches between Fast and Thinking modes. Users can choose from these three modes using the model picker in ChatGPT.

“Starting today, GPT-5 is going to appear smarter. Yesterday, the autoswitcher malfunctioned and was down for a good part of the day, which made GPT-5 look a lot less intelligent”, Altman shared in a post on X.

GPT-4o Is Back

In a Reddit Ask Me Anything session, several users asked OpenAI to reintroduce GPT-4o.

In response to Reddit discussion, Altman mentioned that the OpenAI team had listened to user feedback and decided to provide an option for Plus users to keep using GPT-4o, adding that they would monitor its usage to figure out how long to keep supporting it.

GPT-4o is now listed under “Legacy models” by default for those who pay. Other older AI models like o3 and GPT-4.1, along with GPT-5 Thinking mini, can be included in the model picker in ChatGPT by turning on “Show additional models” in the settings. Just to clarify, this feature is only for paid users.

ChatGPT Plus and Team subscribers can now send up to 3,000 messages weekly when using GPT-5 in Thinking mode, with additional capacity on GPT-5 Thinking mini once they reach this limit.

The company has also made GPT-5 accessible for ChatGPT Enterprise and Edu subscribers.

Moreover, OpenAI has stated that it is working on enhancements at the user interface (UI) level to make it easier for users to activate Thinking Mode in GPT-5 and to better identify which AI model is answering their questions or prompts.