OpenAI’s GPT-5, released on 7 August, has drawn a mixed reception from researchers and customers who expected a step-change over GPT-4o. Independent benchmark tests reported by the Financial Times show only incremental performance gains, while early adopters complained that the chatbot felt curt, lost context and underperformed rival models from Anthropic, Google and xAI. Chief Executive Officer Sam Altman conceded the launch was “a little more bumpy than we’d hoped” and told reporters the company “screwed up” parts of the rollout. In response, OpenAI has restored access to GPT-4o for paid users, adjusted GPT-5’s tone to sound warmer, and introduced Auto, Fast and Thinking modes that let customers trade speed for deeper reasoning. Plus and Team subscribers can now send up to 3,000 messages a week in GPT-5 Thinking mode. The criticism has not dented demand. Altman said API traffic doubled within 48 hours of the release and that ChatGPT is approaching 700 million weekly users. Oracle on Monday became the first major customer to embed GPT-5 across its database, NetSuite and Fusion Cloud applications, citing the model’s code-generation and reasoning abilities. GPT-5 is also available through ChatGPT Enterprise and the OpenAI API in three size tiers. Altman warned that OpenAI is constrained by chip shortages and pledged to double computing capacity in the next five months. Analysts say the episode underscores both the commercial appetite for large language models and the growing technical and financial hurdles to scaling them further.