OpenAI on Thursday unveiled GPT-5, describing it as its most capable large-language model yet and making it immediately available to all 700 million weekly users of ChatGPT, including those on the free tier. The release, streamed from the company’s San Francisco headquarters, comes after more than two years of development and repeated delays. Chief Executive Officer Sam Altman called GPT-5 “the best model in the world,” citing major gains in reasoning, coding and factual accuracy. The system combines fast conversational responses with a deeper-thinking mode that automatically engages on complex queries. In internal tests, GPT-5 scored 74.9 percent on the SWE-bench Verified software-engineering benchmark and cut hallucination rates on health questions to 1.6 percent, down from 12.9 percent for GPT-4o. The model is being rolled out in four flavors—GPT-5, mini, nano and a higher-capacity Pro version—with the ChatGPT interface routing users to the most appropriate tier. Developers can access the family through OpenAI’s API at about $1.25 per million input tokens and $10 per million output tokens; free ChatGPT users will shift to a lighter model once usage caps are reached, while Plus and Pro subscribers receive higher or unlimited allocations. Altman framed the launch as a bid to reassert leadership in a crowded field that now includes rival models from Anthropic, Google DeepMind, Meta and xAI. Backed by Microsoft, OpenAI is seeking to translate record capital spending on AI infrastructure into enterprise demand, touting GPT-5’s ability to generate ‘software on demand’ as well as handle finance and health-care tasks. The company said the model was red-teamed for more than 5,000 hours and uses a new “safe completion” approach designed to provide helpful answers while blocking misuse.