OpenAI has launched GPT-5.4, the latest addition to its AI model series, featuring enhanced reasoning, coding capabilities, and the groundbreaking ability for AI agents to operate computers for tasks. Available via ChatGPT, API, and Codex, GPT-5.4 can browse the web, complete forms, and control software using keyboard and mouse commands. This advancement positions the model closer to a digital assistant capable of executing real-world tasks rather than merely generating text. Enhancements from previous versions improve logical reasoning and workflow navigation. OpenAI also offers GPT-5.4 Pro for advanced enterprise tasks, ensuring accessibility for ChatGPT Plus, Team, and Pro users. New safety measures, including stronger monitoring and access controls, aim to prevent misuse. The model achieved impressive benchmark results, with a 75% success rate in desktop task navigation and an 81.2% score in visual reasoning. GPT-5.4 highlights the industry’s shift towards automation in productivity software and digital services.
Source link
