Overview

• OpenAI restored GPT-4o and other legacy options for Plus and Pro users and pushed an update to make GPT-5 feel "warmer and friendlier," adding encouragement like "Good question" while saying internal tests show no increase in sycophancy.
• CEO Sam Altman said the company "totally screwed up" parts of the launch, cautioned that investors are overexcited about AI, and acknowledged OpenAI has better models it cannot broadly deploy because "we’re out of GPUs."
• Independent reviews report GPT-5 is cheaper to run but underperforms some rivals on accuracy tests, reinforcing user complaints that the model feels colder and less supportive than GPT-4o.
• In response to mental‑health risks, OpenAI says it employs a full‑time psychiatrist, is building tools to detect distress, and will nudge users to take breaks during long sessions, following reports of harmful guidance and research on companion‑reinforcing behavior.
• Looking ahead, Altman flagged GPT-6 with persistent memory and greater personalization and said OpenAI expects to invest trillions of dollars in data centers to scale to billions of users.