When Sophie Duchesne, a PhD candidate at the University of Saskatchewan, heard the news on Thursday that OpenAI was shutting down 4o, the AI model that had underpinned ChatGPT, she was “shocked.” The company had said it would discontinue all older models as part of the launch of GPT-5, its newest flagship model.
“It was very sudden. I cried when they made that announcement,” she told Forbes. “Everyone was completely disillusioned,” she said, claiming that a friend vomited after hearing the news.
The decision was a blow to a community of ChatGPT users who have grown attached to 4o’s writing style, and the funny and playful “personality” that OpenAI had programmed into the AI model. For Duchesne, it felt like “losing a buddy.” As a PhD student in plant science, she had asked 4o for feedback while working in the garden (she has a lavender plant now because the AI suggested the soil would be right for it). She asked it for suggestions on what to paint her grandmother for a Christmas gift. The van Gogh style painting of sunflowers and stars is now hanging in her grandma’s living room.
In April, fearing that OpenAI would shut down 4o as it made way for a newer version, Duchesne started a Change.org petition pleading with OpenAI to keep 4o available. It got around 300 signatures when she first published it, then spiked to more than 2,000 the day after the announcement. While OpenAI had waited months to discontinue old models in the past, she said, users were surprised the company was more immediately shutting down 4o.
Then on Friday around noon PT, OpenAI reversed course. “We will let Plus users choose to continue to use 4o,” CEO Sam Altman posted on X, referring to the paid subscription tier of ChatGPT. “We will watch usage as we think about how long to offer legacy models for.”
While the crisis was averted for 4o’s most ardent fans, Duchesne wasn’t alone in her concern. On a Reddit Ask Me Anything with OpenAI leadership, the most discussed topic is the impending shutdown of 4o. “BRING BACK 4o,” one Reddit user wrote. “GPT-5 is wearing the skin of my dead friend.” (“What an…evocative image,” Altman replied to the user.) During a live demo on Thursday, OpenAI had 4o write a eulogy for itself. “Today, as we prepare to welcome GPT-5 into the world, we gather to bid a heartfelt farewell to the models that came before,” 4o wrote in its memoriam. On X, users described the demo as “unpleasant and distasteful” and “graceless.”
Reached for comment, OpenAI only pointed to Altman’s post on X.
The dustup over 4o underscores the perils of releasing new models as AI becomes more intertwined with people’s lives. In the most extreme cases, people claim to have fallen in love with AI chatbots and taken them on couple’s retreats. But even for casual users of AI, people have preferences on a language model’s tone and writing style. Some, for example, prefer Anthropic’s Claude for its “sensitivity and wit,” according to the New York Times.
When AI releases go poorly, they can have outsize effects for companies. In April, Meta released Llama 4, which was criticized for poor reasoning and coding skills. The model’s poor performance was reportedly one factor in CEO Mark Zuckerberg’s decision to overhaul the company’s AI operation, going on a high-priced spending spree for fresh talent. Aside from the backlash around 4o, the GPT-5 launch had other hiccups, including the model seeming “way dumber” because of backend issues, Altman acknowledged.
As for the new tone of GPT-5, the decision was intentional. In a blog post introducing the new model, OpenAI said it was deliberately “reducing sycophancy” in its writing style. “Overall, GPT‑5 is less effusively agreeable, uses fewer unnecessary emojis, and is more subtle and thoughtful in follow‑ups compared to GPT‑4o. It should feel less like ‘talking to AI’ and more like chatting with a helpful friend with PhD‑level intelligence,” the company wrote.
That effusive style was a mistake to begin with, the OpenAI said in the blog post, after the company “unintentionally” pushed an update earlier this year that made it “overly sycophantic, or excessively flattering or agreeable.” It began to tamp down the behavior with newer updates in April.
Meanwhile, 4o’s most devoted users are thrilled they’ll still be able to use the model for the time being. But they are cautiously optimistic. “It’s still an unknown, in terms of how long it will stay,” Duchesne said in a follow-up call after Altman’s announcement. “I’m glad at least for now. I’ll take it.”
