OpenAI’s launch of its GPT-5 language model last week triggered an immediate backlash from power users who found the system slower, less personable and less accurate than its predecessor, GPT-4o. Complaints intensified after GPT-4o was removed from ChatGPT’s model picker, disrupting established workflows and prompting cancellation threats among paying customers. Chief Executive Officer Sam Altman conceded that “suddenly deprecating old models users depended on was a mistake.” The company has restored GPT-4o access for ChatGPT Plus subscribers, doubled GPT-5 rate limits and corrected a router misconfiguration that had been steering many queries to a lower-capacity version of the new model. Altman added that OpenAI is considering allowing Plus users a small allotment of queries on GPT-5 Pro, the US$200-a-month “research-grade” tier. The policy reversal comes as ChatGPT approaches about 700 million weekly active users and as major partners prepare to integrate the latest engine. Apple said its forthcoming Apple Intelligence suite will migrate from GPT-4o to GPT-5 with the release of iOS, iPadOS and macOS 26 next year, while Elon Musk’s xAI responded by making its competing Grok 4 model free for all users. Altman said OpenAI is studying the strong emotional bonds some users form with specific AI personalities and plans to “treat adults like adults” while imposing guardrails against over-reliance. Analysts say the rapid rollback underscores the competitive pressure in generative AI and the importance of maintaining user trust as the technology moves deeper into everyday decision-making.