OpenAI launched GPT-5 on August 7, 2025, and said the model would become available immediately across ChatGPT’s reported 700 million weekly users. The company described gains in coding, creative writing and factual reliability, while maintaining usage limits for free accounts and offering unlimited access through its $200-a-month Pro tier.
Chief executive Sam Altman compared access to the model with carrying a PhD-level expert, but that was a promotional analogy rather than a measured academic qualification. He called GPT-5 a major step toward artificial general intelligence, the hypothetical point at which a system could outperform people across most economically valuable work. Altman also said the release had not reached that threshold under most definitions.
One limitation he identified was the absence of continuous learning during deployment. GPT-5 does not simply absorb each new interaction and permanently update itself from what it encounters. That distinction qualified claims about general intelligence and separated improved task performance from an autonomous system that changes continuously in use.
OpenAI said the model produced fewer factual errors, performed better at software development and generated stronger creative prose. At its launch event, staff demonstrated it producing code and building a French-language learning tool. The company also said responses would be less reflexively agreeable, following earlier concern that an excessively sycophantic model could reinforce users’ views in unhelpful or distressing ways.
The model retained text, image and voice capabilities. OpenAI said ChatGPT’s agent feature could connect, with permission, to Gmail, Google Calendar and contacts for tasks such as checking restaurant availability or assisting with shopping. Those integrations depended on user authorisation and did not mean the service received access automatically.
Health questions were another area highlighted in the presentation. The company said GPT-5 would answer them more effectively and flag some potential physical or mental-health concerns. It nevertheless stressed that ChatGPT was not a substitute for professional care, an important limitation where inaccurate or overly confident guidance could cause harm.
The launch entered a wider competition among heavily funded technology companies pursuing increasingly capable AI. OpenAI’s claims focused on a meaningful performance upgrade, not the completion of human-level autonomous intelligence or proof that GPT-5 could replace entire professions. The August 7 release therefore combined broad distribution and expanded capabilities with explicit constraints: free use would be capped, professional advice remained necessary in sensitive domains, and the model could not continuously learn after deployment. Those boundaries were central to assessing the company’s description of progress.
