OpenAI chief executive officer Sam Altman acknowledged missteps in last week’s launch of the company’s GPT-5 model, telling reporters the rollout “totally screwed up some things,” citing a severe GPU shortage and unanticipated user backlash. The company has restored access to the earlier GPT-4o model and added a manual “model picker” after complaints that GPT-5 felt less personable and, in some tasks, under-performed rival systems from Google and Anthropic. Altman said the decision to prioritise lower inference costs over raw size was deliberate, but conceded communication with consumers “could have been better.” Early testing presents a mixed picture. Developers interviewed by WIRED and other outlets praise GPT-5’s technical-reasoning speed and lower price, yet several benchmark trackers place it behind Anthropic’s Claude Opus in code generation and below some Chinese models in specialised tasks. The Verge described the release as “overhyped and underwhelming,” while AI critic Gary Marcus called it OpenAI’s “Waterloo.” Financial results tell a different story. According to analytics firm Appfigures, the ChatGPT mobile application has generated about US$2 billion in global consumer spending since debuting on iOS and Android in May 2023. Of that total, US$1.35 billion was booked in 2025 alone, a 673 percent year-on-year surge, with average revenue per install reaching US$2.91. Altman said ChatGPT now approaches 700 million weekly active users and that API traffic doubled within two days of the GPT-5 release, straining the company’s GPU capacity. Looking ahead, Altman outlined ambitions that stretch beyond foundational models: an AI-powered web browser, new consumer apps under incoming applications chief Fidji Simo, possible investments in brain–computer interfaces, and what he described as “trillions of dollars” in future data-centre spending. OpenAI is also evaluating advertising and affiliate commerce inside ChatGPT as additional revenue streams, underscoring how the company hopes to monetise its user base even as it works to repair the reputational damage from GPT-5’s rocky debut.