GPT-5.4 is OpenAI's unified professional AI system, designed to consolidate reasoning, coding, and agentic capabilities into a single cohesive model. One defining capability is its native computer-use functionality, which lets the model interpret screenshots, navigate interfaces, execute commands, and autonomously complete complex workflows—a meaningful step beyond generating text responses. The model can also present an upfront plan before execution, allowing users to review and adjust the approach mid-task. Its architecture absorbs the strong coding capabilities from prior Codex releases while adding improved performance across tools, spreadsheets, presentations, and documents. Multimodal processing handles text, images, and PDFs, and the large context window enables analysis of entire codebases or document collections in single requests.

The GPT-5.4 family spans multiple variants optimized for different performance and efficiency needs. On SWE-Bench Pro, GPT-5.4 scores 57.7%, while its performance on OSWorld reaches 75%—surpassing the 72.4% human expert baseline for desktop computer use. The Mini variant runs more than twice as fast as previous small models while approaching the larger model's results on key evaluations; Nano targets classification, data extraction, and lighter subagent tasks where speed and cost matter most. Available through ChatGPT, the API, and Codex, the model serves as a foundation for building enterprise AI agents that need reliable tool use, coding assistance, and computer automation. The family also includes a Pro variant for maximum performance on complex tasks and a Thinking mode for interactive reasoning.