OpenAI chief executive officer Sam Altman acknowledged flaws in last week’s debut of GPT-5, conceding in multiple media interviews that the company “totally screwed up some things on the rollout.” Users complained that the new flagship model sounded less personable and, at times, delivered poorer answers than its predecessor GPT-4o, forcing the company to restore the earlier system as an option inside ChatGPT. Independent tests and public leaderboards intensified the backlash. On the widely followed lmarena ranking, several GPT-5 variants lagged behind Chinese large-language models such as Kimi-K2 and Qwen3-235B on coding benchmarks, while Ars Technica and The Verge described the upgrade as incremental. AI analyst Gary Marcus called the model “overhyped and underwhelming,” and social-media criticism quickly spread. Altman defended the release, saying GPT-5 was built for lower inference cost and faster responses amid a “big GPU crunch.” He told journalists that API traffic doubled within 48 hours of launch and that ChatGPT continues to hit record daily usage. New features, including an automatic model-switcher, were designed to improve practical utility even if headline performance gains appear modest, he said. Looking ahead, Altman predicted that ChatGPT could soon handle more daily conversations than all humans combined and reiterated plans to invest “trillions of dollars” in data-center capacity to reach artificial general intelligence. He acknowledged that an “AI bubble” is forming but argued the underlying technology remains transformative. OpenAI, which raised about $40 billion earlier this year at a near-$300 billion valuation, is exploring novel financing instruments to fund the next phase of compute expansion.