OpenAI Launches GPT-5 Model

OpenAI has launched GPT-5, its latest AI model, boasting advancements in coding, agentic skills, and contextual understanding. GPT-5 is a unified system that dynamically selects between fast responses and deeper reasoning based on the prompt's complexity. It features improved capabilities in writing, coding, and health-related inquiries. The model demonstrates enhanced performance in coding tasks, complex front-end generation, and debugging. It can create responsive websites, apps and games from a single prompt. GPT-5 scores higher than previous models on benchmarks in maths, coding, visual perception and health.

GPT-5 is available to all users, with increased usage for Plus subscribers and access to GPT-5 Pro, offering extended reasoning, for Pro subscribers. It is integrated into Microsoft's Copilot and is in public preview for GitHub Copilot. Developers can access GPT-5 models in Azure AI Foundry, benefiting from enterprise-grade security and compliance. GPT-5 introduces 'minimal' reasoning and a 'verbosity' parameter in the API. The model supports text and image inputs, with text-only output, and has an input limit of 272,000 tokens and an output limit of 128,000 tokens.

While GPT-5 generally shows modest but significant improvements, some have observed that the model looks very different compared to GPT-4. OpenAI has also faced scrutiny regarding the accuracy of graphs presented during the official launch. Despite this, GPT-5 is considered a significant step towards artificial general intelligence, with potential to surpass humans in economically valuable work.