OpenAI is making swift changes to its latest AI model, GPT-5, following widespread criticism from users over performance and accessibility issues. The backlash erupted soon after GPT-5’s rollout, with complaints centering on a model-switching glitch that frequently routed conversations to older, less capable versions, making the system appear “dumber” than expected.

CEO Sam Altman acknowledged the rocky launch and confirmed that GPT-4o, a popular earlier model praised for its personable tone, has been reinstated for ChatGPT Plus subscribers at US$20 per month. The decision came just a day after GPT-5’s debut, following strong user demand for the option to switch back.

OpenAI is also addressing technical concerns by working on a more reliable model selection system, doubling GPT-5’s usage rate limits for paying users, and introducing a dedicated “thinking mode” for complex tasks.

The update brings new personalization features, with GPT-5 offering four selectable “personalities”: Cynic, Robot, Listener, and Nerd, alongside improved coding tools, reduced hallucinations, and tighter integration with Gmail, Google Calendar, and a new Study Mode.

In a move to enhance safety, GPT-5 will now avoid providing direct emotional or life-altering advice, instead encouraging self-reflection and responsible decision-making. This change aligns with OpenAI’s evolving ethics-first approach to AI interactions.

Industry observers note that the company’s strategy of releasing features early to gather real-world feedback played a major role in shaping these rapid adjustments. Despite early setbacks, OpenAI says it remains committed to refining GPT-5’s performance and restoring user confidence.