OpenAI on Thursday unveiled GPT-5, calling the system its “smartest, fastest, most useful model yet” and immediately rolling it out to all 700 million weekly ChatGPT users, including those on the free tier. Chief Executive Officer Sam Altman described the release as “a significant step toward AGI” and “the best model in the world for coding and writing.” It is the first time OpenAI’s flagship model has been made available without a subscription, following more than two years of development marked by delays and public speculation. GPT-5 ships in four variants—standard, mini, nano and the higher-latency “Thinking” version—and is fronted by an automatic router that decides when deeper reasoning is required. OpenAI says the model cuts hallucinations by roughly 45 % compared with GPT-4o, expands the context window to 400,000 tokens and sets new highs on benchmarks covering math, coding, multimodal reasoning and health queries. New safety training emphasises “safe completions” over blanket refusals, while optional preset personalities and an upgraded voice mode aim to make interactions feel more natural. Free ChatGPT users are limited to 10 GPT-5 messages every five hours plus one daily Thinking request; Plus subscribers receive 80 messages every three hours, and Teams or Pro customers have unlimited access. Enterprise and education accounts are scheduled to join next week. In a parallel move, Microsoft integrated GPT-5 across its Copilot suite—including Microsoft 365, GitHub Copilot and Azure AI Foundry—broadening the model’s reach to business and developer workflows on the day of launch.