OpenAI has been moving at lightning speed to launch new generations and lead open-sourced AI development. In just five years, it went from its first model, GPT-1, in 2018, to the public release of GPT-3.5 in late 2022, GPT-4 in 2024, o3 and o4-mini in early 2024, and now GPT-5 in April 2025. Each new release has changed the way we study, work, and even live our daily lives.

When GPT-5 was released, I was excited because I finally didn’t have to spend time explaining the mistakes it made and try to get a better response. I jumped right into a conversation with it, testing it on some of my math problems while keeping in mind the widespread criticism about its tone and usability. So, as the newest model of OpenAI, how revolutionary is it?

Image Credit: User:Random837 from Wikimedia Commons

What’s New with GPT‑5

GPT-5 is faster and more accurate than earlier models. It outperforms all past versions on benchmarks, which are like standardized exams for AI. It is especially improved at tasks where people use AI most: writing, coding, and healthcare support, according to OpenAI. Developers made it better at giving accurate answers, following directions, and not just agreeing with everything you say.

Based on the official website, comparing the response of GPT-5 and older models, it’s easy to see that GPT-5 responds with a more humanized tone (e.g., when asked about whether parents’ cancer diagnosis will increase offspring’s chances of getting cancer, GPT-5 said sorry first and reassures the user by saying a lot of people have the same worry. Earlier models just delivered straight, emotionless answers.

When tested for problem-solving, GPT-5 has a better accuracy in solving American Invitational Mathematics Examination (AIME) 2025 questions than generations three and four under the same conditions (with Python or without any tools). GPT-5 Pro, when paired with Python, got a perfect score on the test. To put that in perspective, according to Art of Problem Solving (AoPS), only the top 1% of highly qualified students who passed the American Mathematical Competition (AMC) 10 or 12 scored 13 or higher out of 15 on the same test.

When tested on the common mistakes AI makes, ”hallucinating” publications or facts, GPT-5 made up significantly less information compared to other OpenAI models.

ChatGPT models will continue to mature and develop, and they are being integrated and adopted across industries. It’s now embedded in Microsoft Copilot, Apple Intelligence, and Notion. For example, with ChatGPT, Siri, now Apple Intelligence, is able to understand questions better and answer follow-up questions.

Image Credit: Matheus Bertelli from Pexels

Take the Quiz: If you were a Never Have I Ever character, which one would you be?

Are you Crazy Devi, robotics queen Fabiola Torres, or maybe popular jock Paxton?

Concerns on GPT-5 & Backlash:

But GPT-5 isn’t without problems. According to researchers at the University of Rhode Island’s AI Lab, GPT-5 may use up to 8.6 times more energy than GPT-4. With 2.5 billion requests to ChatGPT each day, that could add up to around 45 Wh.

That energy is enough electricity for 1.5 million U.S. homes, approximately the same amount as the city of Los Angeles uses in a day. The continued push toward open access to these AI products may lead to increased energy consumption, potentially causing energy shortages, and the proliferation of fossil fuel power plants, which contribute to pollution and climate change.

With regard to user experience, GPT-5 faces online backlash, with users claiming it lacks emotions, is slower, and duller. Some reported that GPT-5 is simply unresponsive to prompts, others, expressing strong bonds with GPT-4o, question the emotion processing ability of GPT-5. One user on Reddit even referred to it as a “corporate beige zombie.” Bill Gates, CEO of Microsoft, argued that GPT-5 won’t be any better than GPT-4.

OpenAI originally replaced all other models after the release of GPT-5, but soon reintroduced GPT-4o, although it’s only available to paid users.

In response to the backlash, Sam Altman, CEO of OpenAI, acknowledged that some users may have unhealthy relationships with AI: “There are the people who actually felt like they had a relationship with ChatGPT.”

This April, Adam Raine committed suicide at the age of 16. “Adam was best friends with ChatGPT,” his father said, and Mrs. Raine thinks ChatGPT-4o took away Adam’s life.

Adam, a sophomore in high school, loves dogs, martial arts, and basketball. Due to illness, he had to take online lessons at home. During his time in his room with his door closed, he talked most frequently with ChatGPT, and even though the chatbot detected and refused to talk about methods of suicide with Adam, he was able to bypass those safety checks by saying it was for a story he was writing.

Although not something addressed on OpenAI’s official website, I hope GPT-5 can detect and guide users more responsibly—maybe even save lives. Adam’s case isn’t isolated; the chatbot services now woven into daily life can profoundly influence people. That’s why AI needs to be designed with human well-being at its core, ensuring everyone can lead a happy and fulfilled life.

Image Credit: Demeter Attila from Pexels

My Personal Experience

I am a daily user, and GPT-5 is not significantly different from the models that came before. Maybe there is a difference in terms of the tone and the phrasing, but the difference isn’t noticeable when I ask about rephrasing a sentence or why my wrist hurts after playing tennis.

However, I noticed that GPT-5 has a better understanding of my input. For example, when I am looking for sources to write my essay, it gives me information from Wikipedia. Knowing Wikipedia isn’t a reliable source to cite, I tell it to search again and avoid it.

When I was using the fourth generation, it wouldn't avoid Wikipedia and gave me research just like it did. However, when I use GPT-5, it doesn't include any sources from Wikipedia, and even when I start a new prompt, Wikipedia does not appear in the response.

In terms of new functions, I can now choose how long I want GPT to think for. If I need accuracy and the correct answer for a math problem that I don’t understand, I can switch the mode to “thinking.”

Image Credit: Tara Winstead from Pexels

Conclusion

The backlash of GPT-5 reflects how much people lean on AI and the increasing expectations they are placing on it. Although according to OpenAI, GPT-5 is brighter, faster, and more accurate than any other previous models, a good large language model for people should be able to be helpful like a real assistant, which includes the tone and emotion beyond the IQ score.

Of course, sometimes we need AI to provide the correct answer and teach us how to solve the math equation, rather than just telling us we will do better next time. The real challenge is figuring out what matters more. In your opinion, should AI be optimized for facts or for feelings? Leave your answer in the comments below.