OpenAI on 7 August released GPT-5, its new flagship language model that now powers ChatGPT, calling it the company’s “smartest, fastest and most useful” system to date. Chief Executive Officer Sam Altman likened the model’s expertise to “Ph.D-level” intelligence and said the upgrade represents the biggest single step in making advanced AI broadly accessible. GPT-5 improves logical reasoning, memory and multimodal understanding of text, images, audio and video. Internal benchmarks show the model reduces factual errors by 45% compared with GPT-4o and lifts coding accuracy to 74.9% on the SWE-bench test. A high-end “GPT-5 Thinking” variant offers a 256,000-token context window for complex tasks, while a router automatically chooses faster or deeper reasoning modes as needed. The model is available immediately to all ChatGPT users worldwide, including the free tier, with higher usage caps for the $20-per-month Plus plan and an unlimited Pro option. Enterprise and Education customers will gain access next week. OpenAI has also introduced native connectors that let paid users search files in Dropbox, Microsoft Teams, Gmail and other services without leaving ChatGPT. A surge in demand exposed capacity constraints and prompted user backlash over the removal of older models. Within days, OpenAI restored the popular GPT-4o for Plus subscribers, doubled GPT-5 message limits and pledged to add a mid-priced plan for heavy individual users. Altman acknowledged the launch was “bumpier than we hoped” and said the company will double its compute fleet over the next five months to meet load. With 700 million people now using ChatGPT each week, GPT-5 intensifies competition with rivals such as xAI’s Grok, Google’s Gemini 2.5 and Anthropic’s Claude 4.1. By offering the new model free at launch while emphasizing stronger coding, lower hallucination rates and tighter third-party integrations, OpenAI aims to defend its lead in the fast-moving generative-AI market.