GPT-5.4: A Unified AI Model Built for Real-World Work
GPT-5.4 unifies world-class coding, native computer use, million-token context, and deep reasoning into a single, blazingly efficient model. Built for the way knowledge workers actually work.
Trusted by users from 10,000+ companies
GPT-5.4 introduces a range of improvements designed to support advanced reasoning, automation, and software development.
Uses structured reasoning to solve complex problems, analyze information, and plan multi-step workflows with higher accuracy.
Combines powerful coding capabilities with strong general knowledge reasoning to handle technical and creative tasks.
Processes massive inputs using a context window nearing one million tokens across documents, datasets, and codebases.
Interacts with software interfaces using mouse actions, keyboard commands, and automated testing workflows like humans.
Analyzes images, screenshots, and visual layouts to interpret designs, data, and interface structures accurately.
Works seamlessly with browsers, development environments, and productivity tools to complete complex digital workflows.
Checks its own outputs, runs validation steps, and improves results through automated testing and analysis.
Optimizes planning and tool usage to reduce token consumption during long workflows and complex tasks.
Creates clear execution plans before performing tasks to guide workflows and reduce unnecessary processing.
GPT-5.4 was designed to outperform previous models in tasks that reflect real-world productivity.

On the GDP-Eval benchmark, which measures a model’s ability to perform practical knowledge tasks, GPT-5.4 Thinking achieved an impressive 83% score, outperforming previous models and competing systems.

GPT-5.4 improves software engineering performance with a 57.7% score on SWE-Bench Pro, exceeding the results of earlier coding-focused models. This demonstrates its ability to debug code and handle complex development workflows more reliably.

In the OSWorld benchmark, which measures an AI’s ability to interact with computer interfaces, GPT-5.4 reaches around 75% accuracy with significantly fewer tool calls. This means faster workflows and lower token costs.
GPT-5.4 can analyze reports, summarize large documents, and generate structured spreadsheets automatically. This makes it valuable for research, financial analysis, and business operations.
