OpenAI on Thursday unveiled GPT-5, a new flagship language model that the company says delivers PhD-level reasoning across coding, writing, mathematics and health queries. Chief Executive Officer Sam Altman called the system “the best model in the world,” describing the release as a significant step toward more general forms of artificial intelligence. GPT-5 introduces a unified architecture that automatically routes complex prompts to a deeper-thinking version of the model, while simpler questions are handled by a faster core engine. OpenAI says the system’s responses are about 45% less likely to contain factual errors than GPT-4o and, when the optional “thinking” mode is engaged, hallucinations drop by roughly 80%. The model supports a context window of up to 256,000 tokens and ships alongside smaller GPT-5 mini and nano versions, as well as a higher-capacity GPT-5 pro variant. The software is rolling out immediately inside ChatGPT. Free users receive up to 10 GPT-5 messages every five hours plus one daily “thinking” request before the service falls back to a lighter model. ChatGPT Plus subscribers are allotted 80 messages every three hours, while Pro and Team tiers get unlimited access and can invoke GPT-5 pro for extended reasoning. OpenAI says Enterprise and Education customers will follow within a week. Developers can tap the model through OpenAI’s API at $1.25 per million input tokens and $10 per million output tokens, with lower-cost options for the mini and nano versions. The launch comes as ChatGPT approaches 700 million weekly users and intensifies competition with rival systems from Google, Anthropic and Elon Musk’s xAI.