OpenAI unveiled GPT-5.4, optimized for professional tasks, on the 6th.
GPT-5.4 is a next-generation model applied across OpenAI's major products, including ChatGPT, the application programming interface (API), and Codex. It integrates reasoning ability, coding performance, and agent-based workflows into a single model.
By integrating the coding capabilities of GPT-5.3-Codex, GPT-5.4 improves how various tools and software are used in professional environments such as spreadsheets, presentations, and documents. It is designed to reduce repetitive tasks and perform complex work more efficiently.
In terms of performance, it matched or exceeded industry experts on 83% of tasks in GDPval, a benchmark that evaluates real knowledge-based task execution. That is an improvement over GPT-5.2's 71.0%.
In internal spreadsheet modeling evaluations, it scored an average of 87.5%, surpassing GPT-5.2's 68.4%. Presentation creation also saw improvements in design completeness, visual diversity, and factual accuracy.
Among general-purpose models, GPT-5.4 is the first to come standard with computer-use functionality. In Codex and API environments, agents can manipulate software in real computer environments and carry out tasks across multiple applications. It supports up to a 1 million-token context, making it useful for long-duration planning and execution.
GPT-5.4 provided in ChatGPT also introduces a method that presents a work plan before generating an answer through the "GPT-5.4 Thinking" mode. It further strengthens web-based deep research to improve response accuracy for questions that synthesize multiple sources.
In addition, the tool search feature makes it possible to select and use the necessary tools in environments where various tools are connected. OpenAI said GPT-5.4 reduces token usage compared with previous models while delivering efficient reasoning performance.