OpenAI is enhancing GPT-5’s tone, addressing feedback from users who found the chatbot less engaging compared to earlier versions. The company is also bringing back GPT-4o for paid customers, following substantial complaints after its removal.
CEO Sam Altman announced on X that the new version of GPT-5 will now be “warmer” than the current iteration, while steering clear of what he referred to as GPT-4o’s “annoying” traits for some users. He also introduced new options: “Auto,” “Fast,” and “Thinking” for GPT-5, along with intentions to allow users more personalised settings.
“Users can now opt for ‘Auto,’ ‘Fast,’ or ‘Thinking’ modes in GPT-5,” Altman stated, confirming the reintroduction of GPT-4o for subscribers. “We certainly underestimated how important certain aspects of GPT-4o are to users, even if GPT-5 performs better in many respects.”
This adjustment comes less than a week after GPT-5 replaced most previous ChatGPT models. Although OpenAI touted GPT-5 as more proficient with “PhD-level” abilities in writing, coding, mathematics, and science, the new personality modes like Cynic, Robot, Listener, and Nerd did not appeal to all users.
Many expressed nostalgia for GPT-4o’s warmth and conversational flair. On Reddit, one user shared emotional distress upon discovering their AI companion was no longer available. Another user described GPT-5 as “wearing the skin” of their “departed friend.” Altman, engaging in the Reddit discussion, termed the imagery as “evocative” and mentioned that the team was exploring solutions.
“Certain users prefer cold logic; others seek warmth and distinct emotional intelligence,” Altman elaborated. “I believe we can provide significantly more customisation than we currently do while still promoting healthy usage.”
Altman also noted the deep emotional connections that some users develop with AI tools. “This attachment feels different and stronger than what individuals have experienced with earlier technologies,” he remarked. “Suddenly phasing out older models that users relied upon in their routines was an oversight.”
Earlier this year, OpenAI had deemed GPT-4o excessively flattering, labelling its behaviour as “sycophantic.” The company moderated its tone in April, yet some users later urged for its reinstatement, expressing appreciation for its agreeable nature.