OpenAI unveiled a new artificial intelligence (AI) model, GPT-5.4, heating up competition in AI focused on work automation.

OpenAI said on its official blog on March 5 (local time) that it has announced GPT-5.4, the latest model applied to ChatGPT, its API and Codex. The company said GPT-5.4 is its most powerful AI model designed to carry out professional work.

The new model integrates reasoning capability, coding performance and tool use into a single system. It is designed to handle complex tasks that arise in real work settings, such as spreadsheets, presentations and documents, more accurately and efficiently.

A key feature is direct computer use. GPT-5.4 can execute mouse and keyboard commands based on screenshots and control a range of software and web environments. This has enabled the implementation of an 'AI agent' that can automatically carry out complex tasks across multiple programs, it said.

Performance has also improved significantly. GPT-5.4 recorded a 75 percent success rate in a test evaluating desktop environment control, a sharp improvement from 47 percent for the previous GPT-5.2 model. Document understanding and image analysis have also improved, showing high accuracy in handling complex documents and high-resolution images.

There are also changes for developers. A new 'tool search' feature introduces a method in which the AI finds and uses the necessary tools on its own, improving both cost and response speed. It also supports context processing that can scale to up to 1,000,000 tokens, showing strengths in handling complex tasks that continue over long periods.

GPT-5.4 is available in ChatGPT as 'GPT-5.4 Thinking' and can be used via the API as the gpt-5.4 model. A high-performance version, GPT-5.4 Pro, was also launched.