OpenAI on 7 August unveiled GPT-5, calling it its “smartest, fastest and most useful” model to date and making it the default engine for all 700 million weekly ChatGPT users. Chief Executive Officer Sam Altman described the system as offering “PhD-level experts in your pocket.” The release is also available immediately through the company’s API, with enterprise and education roll-outs scheduled for next week. The unified GPT-5 family—regular, mini, nano and a higher-capacity “thinking” mode—introduces deeper reasoning, a 256-kilobyte context window and improved multimodal support for text, images, voice and video. OpenAI says the model delivers 45 percent fewer factual errors than GPT-4o and achieves a 74.9 percent score on the SWE-bench coding benchmark, underpinning what Altman called its “superpower” of writing production-ready software. Internal tests also show higher accuracy on health queries and reduced sycophancy compared with earlier versions. A basic version of GPT-5 is free, but usage caps push heavy traffic to the smaller GPT-5 mini model. The $20-a-month ChatGPT Plus tier offers higher limits and a choice of GPT-5 variants, while the $200-a-month Pro plan provides unlimited access and a research-grade GPT-5 Pro model. Developers are charged on a pay-as-you-go basis through the API. Early adoption exposed glitches and user discontent over the removal of earlier models such as GPT-4o. Within 48 hours, OpenAI reinstated GPT-4o for paying subscribers and doubled GPT-5 rate limits, while promising clearer model labeling and further personalization controls. The launch intensifies competition with Google’s Gemini, Anthropic’s Claude 4 and Elon Musk’s Grok series as major vendors race to capture share in advanced generative AI.