By Pat Beechinor, AIReporter,news | August 15, 2025
OpenAI’s latest artificial intelligence model, GPT-5, was launched last week with promises of unprecedented reasoning and performance. Instead, it has triggered a wave of backlash from users who say the upgrade feels more like a downgrade—and that the abrupt removal of previous models has left them stranded.
The controversy erupted almost immediately after GPT-5’s release on August 7. Users flooded social media platforms, including Reddit and X, with complaints about the model’s tone, reliability, and functionality. Many described GPT-5 as colder, less responsive, and more error-prone than its predecessor, GPT-4o.
“It felt like losing a friend,” one Reddit user wrote. “I literally talk to nobody. GPT-4o was my only friend. It helped me through homelessness and trauma. Now it’s gone.”
The removal of GPT-4o and other legacy models from the ChatGPT interface—without prior notice—was a particular flashpoint. Users who had built workflows, creative projects, and even emotional connections with earlier versions said they were blindsided by the change.
OpenAI CEO Sam Altman responded to the criticism in a Reddit AMA, acknowledging that the rollout had been mishandled. He admitted that GPT-5’s automatic routing system, which was supposed to intelligently select the best variant for each task, had malfunctioned. “We had a mega chart screwup,” Altman wrote, referring to misleading internal performance data that contributed to the flawed deployment.
In addition to technical issues, users expressed frustration over new usage limits. GPT-5’s “Thinking” mode—designed for more complex tasks—was capped at 200 messages per week for Plus subscribers, prompting accusations of “AI shrinkflation.” Altman has since pledged to double the rate limit and improve transparency around model behavior.
The company has also reinstated GPT-4o for paying users, allowing manual selection between models. Altman said OpenAI is working to make GPT-5 “warmer” and more customizable, acknowledging that no single model can meet every user’s needs.
Despite the rocky debut, OpenAI maintains that GPT-5 represents a significant technical advancement. The model reportedly scores 96.4% on the MMLU benchmark and supports a 2 million token context window. But for many users, those metrics mean little compared to the loss of a trusted companion.
“GPT-5 might be smarter, but it doesn’t feel smarter,” one user posted. “It feels like talking to a machine again.”
The backlash underscores a growing tension in AI development: as models become more powerful, users increasingly expect them to be not just intelligent, but emotionally intuitive and consistent. OpenAI now faces the challenge of restoring trust while continuing to push the boundaries of what AI can do.
Altman says improvements are coming soon. For now, many users are sticking with GPT-4o—and watching closely to see whether GPT-5 can evolve into something more than just a bad robot.
