OpenAI has reinstated its popular GPT-4o model for paying ChatGPT users after backlash over the rollout of GPT-5. The decision marks a swift reversal just days after the company removed the option to use legacy models.
CEO Sam Altman hosted a Reddit “Ask Me Anything” (AMA) to discuss GPT-5’s launch, but the thread quickly filled with complaints about the new model’s tone and performance compared to GPT-4o. Altman acknowledged the concerns and later announced on X that subscribers would once again be able to select the older model.
GPT-4o struck a chord with loyal users
During the AMA, subscribers described GPT-4o as a model that matched their communication style, complemented their workflows, and, for some, felt like a familiar companion. Part of its appeal, they said, was a conversational style that could feel supportive or even complimentary, a quality some felt was missing in its replacement.
By contrast, GPT-5 was portrayed as technically sharp but lacking the warmth and personality that had kept users coming back. As part of its launch, OpenAI replaced the model picker with an “autoswitcher” that routed prompts to different GPT-5 variants based on the task, removing direct control and, for some users, breaking carefully built workflows.
Several commenters said this disrupted their routines, while others announced they had canceled their subscriptions in protest.
Altman maintained that the newest model was built to surpass 4o and earlier versions in every respect, highlighting improvements in reasoning and writing quality. Many in the thread disagreed, calling for 4o’s return alongside newer models and the restoration of past perks such as unlimited usage.
Adjustments after the backlash
In response, Altman told Reddit users that GPT-4o would be restored for Plus subscribers and said usage patterns would guide how long it remains available. He also acknowledged requests for access to multiple legacy models and agreed there “should have something unlimited” for heavy users.
Altman admitted on X that OpenAI had underestimated how much users valued certain traits in GPT-4o, even if GPT-5 outperformed it in most ways. He confirmed the change was live through a “show legacy models” setting and said GPT-5 rate limits for Plus users would be doubled.
He also announced they were testing a higher cap of 3,000 reasoning requests per week. The CEO said the autoswitcher had been down for part of the rollout, so the model “seemed way dumber.”
Altman added that adjustments to the decision boundary should help ensure users get the right model more often.
Promises for what’s next
In a follow-up post on X, Altman promised clearer model labeling in the interface and easier ways to trigger thinking mode. He also said near-term changes would make GPT-5’s tone warmer.
In the longer term, he said, OpenAI is investing in “steerability” to let users adjust the system’s tone and emotional intelligence to their preferences, whether that’s adding warmth, removing emojis, or choosing cold logic.
He also warned of capacity constraints in the week ahead, pledging transparency about trade-offs as the rollout continues. GPT-4o’s return leaves GPT-5 as the default while restoring the choice many users demanded, with more updates promised to steady the launch and rebuild subscriber trust.
Researchers say one AI firm has outpaced industry giants in retaining talent, with four out of five hires from 2021 to early 2023 still on the payroll two years later. Find out which company.