OpenAI released GPT-5.4 on March 5, the company's most capable frontier model yet, targeting enterprise customers with a 1 million token context window — the largest in OpenAI's lineup — and native computer-use capabilities that allow the model to autonomously operate applications on behalf of users. It is the first general-use OpenAI model with built-in computer use.

The new model is 33% less likely to make false claims compared to GPT-5.2, with full responses 18% less likely to contain errors. GPT-5.4 scored a record 83% on OpenAI's GDPval benchmark for knowledge work tasks and is significantly more token-efficient than its predecessors.

OpenAI also launched ChatGPT for Excel and Google Sheets in beta, plus enterprise data integrations with FactSet, MSCI, Third Bridge, and Moody's. The model is available in GPT-5.4 Pro and GPT-5.4 Thinking versions for ChatGPT Plus, Team, Pro, Enterprise, and Edu subscribers, as well as through the API.