OpenAI GPT-5 model faces backlash, company brings back legacy models for customers

OpenAI has reinstated the GPT-4o model for paid ChatGPT subscribers, just a day after replacing it with its newly launched GPT-5. The move follows widespread criticism from users who were unhappy with losing access to the older model, which many had integrated into their daily workflows.

The rollout of GPT-5, which OpenAI describes as its most advanced large language model to date, was met with mixed reactions despite promises of significant improvements in writing and coding capabilities. In an AMA session on Reddit, several users told CEO Sam Altman that GPT-5 felt “dumber” and lacked the warmer personality of GPT-4o. Some also expressed frustration at the abrupt removal of legacy models without prior notice.

One Reddit user described the difference as jarring, saying GPT-4o “genuinely talked to me” but GPT-5 responded with “cut-and-dry corporate” replies. Others complained that the company’s new real-time router, designed to select the best model for each task, was not working as intended. Altman acknowledged the issue, confirming that the router had been underperforming and stating that GPT-5 would “seem smarter starting today”. He also pledged to make it clearer which model is responding to a given query.

Initially, the update also removed ChatGPT’s model picker, which previously allowed users to manually select between different versions like GPT-4o, GPT-4.1 or the o4 mini model. As of now, ChatGPT Pro, Team and Enterprise customers can once again choose GPT-4o through the “show legacy models” option in settings, while free-tier users remain limited to GPT-5.

Altman said OpenAI will monitor usage before deciding how long to support older models, and indicated that GPT-4.1 could return if there is enough demand from Plus subscribers. He added that the company will “continue to work to get things stable” and keep listening to user feedback.