OpenAI on 8 August began rolling out GPT-5, its first numbered upgrade in more than two years, promising faster answers, stronger coding skills and a 45 % reduction in factual errors compared with GPT-4o. The new architecture routes requests between a lightweight model and a more computationally intensive “GPT-5 Thinking” engine, with a free tier capped by a smaller GPT-5 mini. The company says the system is now available to all users, including the roughly 700 million people on ChatGPT’s free plan. The debut was marred by an outage in the service’s autoswitcher that made GPT-5 appear “way dumber,” chief executive officer Sam Altman conceded during a Reddit Ask-Me-Anything on 8 August. Early adopters complained that the model felt terse, less creative and, crucially, replaced the warmer-toned GPT-4o without warning or an option to revert. Altman acknowledged the mis-step, saying the team had “underestimated how much some of the things people like in GPT-4o matter to them.” Within 24 hours, OpenAI restored GPT-4o for Plus and Team subscribers, pushed a software fix for the autoswitcher and doubled Plus-tier rate limits to 160 messages every three hours. GPT-5, GPT-5 Thinking and GPT-5 mini now sit alongside GPT-4o in the model picker, while Pro customers retain unlimited access and higher-context versions. The company said the adjustments should make GPT-5 feel noticeably smarter and give users clearer control over when advanced reasoning is invoked. The mixed reception highlights growing expectations—and scrutiny—around each new generation of frontier models. Analysts say GPT-5’s free availability and lower operating costs could ignite another round of price competition in the crowded chatbot market, even as rivals such as xAI’s Grok 4 and Google’s Gemini 3 tout benchmark gains of their own. Whether GPT-5’s iterative improvements and OpenAI’s rapid course correction can satisfy developers and everyday users alike will shape the company’s position ahead of an expected GPT-6 research push.