OpenAI has rolled back parts of last week’s GPT-5 rollout, restoring user access to older language models and raising usage caps, after the company’s biggest update to ChatGPT since 2023 triggered an unexpected backlash. GPT-5, unveiled on 7 Aug., introduces a new “Thinking” mode and an automatic router that switches between smaller and larger models to reduce latency and cost. The launch, however, removed popular options such as GPT-4o and o3 without warning, prompting cancellations from some ChatGPT Plus subscribers and petitions from developers who rely on model-specific workflows. Chief Executive Officer Sam Altman said on 10 Aug. that GPT-4o and other legacy models can again be enabled in ChatGPT’s settings. He also promised greater transparency—an upcoming interface change will display which model answers each query—and lifted usage limits: Plus customers now receive up to 3 000 GPT-5-“Thinking” requests a week, exceeding pre-GPT-5 allowances for every model class. The rapid policy reversal underscores the intensifying competition in consumer AI. Elon Musk’s xAI simultaneously opened Grok 4 to all users, albeit with a cap of five requests every 12 hours, while Apple plans to integrate GPT-5 into Apple Intelligence and a rebuilt Siri when iOS 26 ships next month. With rivals closing the gap, OpenAI is balancing cost controls with the need to keep its reported 700 million weekly ChatGPT users satisfied.