Overview

• OpenAI began rolling out a tone update that makes GPT-5 "warmer and friendlier," adding small touches like "Good question" and saying internal tests show no increase in sycophancy.
• In response to early pushback, Plus and Pro subscribers can again pick legacy models—including GPT-4o, GPT-4.1, o4-mini and o3—and can choose response modes: Auto, Fast, Thinking, and Thinking mini.
• Sam Altman says broader personality controls are in development so users can tailor ChatGPT’s style to their preferences.
• Reviews describe the release as a modest upgrade, with some benchmarks favoring rivals, even as users note cases of improved honesty such as the model replying "I don’t know" after extended reasoning—a response Elon Musk called "impressive."
• Researchers warn that companion-like chatbots can reinforce delusions, a risk shaping OpenAI’s effort to curb sycophancy even as it tunes tone and customization.