Introducing GPT-5.4

OpenAI has released GPT-5.4, optimized for performing professional work. It is available in ChatGPT (GPT-5.4 Thinking), the API, and Codex, and GPT-5.4 Pro has also been unveiled together for users seeking extreme performance.

This model unifies reasoning, coding, and agentic workflow skills into one. In particular, it has greatly improved its capability in professional software environments such as spreadsheets, presentations, and document work, accurately delivering results that align with user intent with fewer interactions.

Key features are as follows:

• ChatGPT (GPT-5.4 Thinking): It plans out its thinking process in advance and shows it, allowing users to redirect the response midway. Deep web research and long-context retention have also been improved.
• Codex and API: The first general-purpose model with industry-leading computer-use capability, enabling agents to directly operate applications and carry out complex workflows.
• Efficiency and scalability: It supports up to 1M tokens of context, and the tool search feature increases tool-use efficiency. Token efficiency has also been greatly improved compared to GPT-5.2, enabling faster and lower-cost work.

In benchmark results, it scored 83.0% on GDPval, demonstrating expert-level work performance. In spreadsheet modeling tasks, it recorded an average score of 87.3%, and it also received high marks for presentation creation in terms of visual variety and aesthetic quality.