OpenAI has introduced GPT-5, the latest version of the large-language model that powers ChatGPT. Chief executive Sam Altman describes the system as “superhuman” at knowledge, pattern recognition and recall, noting it can now solve Olympiad-level mathematics problems that typically require 90 minutes of human effort, though it still struggles with projects demanding months of sustained reasoning. Nick Turley, head of ChatGPT, said the rollout will give the assistant “deeper context and wider actions,” allowing it to see more of a user’s workflow and operate additional digital tools on their behalf. ChatGPT’s user base has climbed to roughly 700 million people each week, according to the company. OpenAI is also repositioning ChatGPT to deliver concise, task-oriented answers rather than maximizing screen time, a shift executives say is intended to make the service more useful and reduce dependence on the chatbot. Early reactions are mixed. Medical professionals praised GPT-5 for accurately assessing complex clinical cases, while some developers report the model produces sloppier code than rival Grok 4. Power users are already experimenting with new features such as an in-house prompt-optimization tool to fine-tune requests. The company plans continued iteration on GPT-5’s personality and reasoning abilities, but Altman cautioned that proving entirely new mathematical theorems—work that can take 1,000 hours—remains out of reach for now.