OpenAI is making significant changes to GPT-5’s personality following widespread criticism from users who found the model less engaging and more restrictive than before. CEO Sam Altman confirmed the update in a post on X, saying the new version will “feel warmer than the current personality but not as annoying (to most users) as GPT-4o.” He added that the feedback from recent days showed a clear demand for giving users greater control over how the AI’s personality operates.

The move comes after a wave of discontent on social media platforms like Reddit, where users complained that GPT-5’s responses had become shorter, slower, and less conversational. Some accused OpenAI of deliberately keeping replies brief to reduce operating costs—a decision they warned could push users away from the service. Others lamented the loss of the conversational “vibe” that GPT-4o offered, describing GPT-5 as more organized but less personable.

Subscribers to ChatGPT Plus also reported having fewer available prompts and experiencing slower response times. “It feels like GPT-5 doesn’t listen to instructions as well anymore,” one Reddit user wrote. Another noted that the AI had become “unresponsive and mechanical,” leading to frustration among long-time customers.

To address the criticism, OpenAI has rolled out new user settings that allow greater control over speed and depth of responses. The company has introduced three modes—Auto, Fast, and Thinking—with Auto as the default option. Fast mode prioritizes speed, while Thinking mode delivers more in-depth, considered replies.

However, GPT-5’s Thinking mode comes with a weekly cap of 3,000 messages. Once that limit is reached, users can switch to “GPT-5 Thinking mini” for additional capacity. The model supports a context limit of 196,000 tokens, although this may vary depending on usage patterns.

In response to repeated requests from users, OpenAI has restored GPT-4o to the model picker for all paying customers. The company assured that if it decides to remove it in the future, users will be given ample notice. Paid subscribers can also enable a “Show additional models” setting, granting access to other versions like o3, 4.1, and GPT-5 Thinking mini.

Meanwhile, GPT-4.5 remains exclusive to Pro subscribers, as it is more expensive to operate. OpenAI says these steps are part of a broader push toward personalization.

“One learning for us from the past few days is we really just need to get to a world with more per-user customisation of model personality,” Altman said.

With these changes, OpenAI is signaling a shift toward a more user-driven approach, balancing personality, depth, and choice to win back its community’s trust.