GPT-5.4

About GPT-5.4

GPT-5.4 is OpenAI's most capable and efficient frontier model for professional work, released in March 2026. It unifies reasoning, coding, and agentic capabilities into a single architecture, integrating the coding strengths of GPT-5.3-Codex with enhanced computer use and knowledge work features. The model operates across ChatGPT as GPT-5.4 Thinking, the API as gpt-5.4, and Codex for software development workflows.

The model introduces native computer use capabilities, enabling agents to operate computers via screenshots and keyboard commands or through libraries like Playwright. It supports up to 1 million tokens of context for long horizon task planning and execution. GPT-5.4 also implements tool search architecture that reduces token usage by approximately 47 percent when working with large tool ecosystems compared to previous approaches.

GPT-5.4 delivers measurable improvements in professional tasks, achieving 83 percent on GDPval benchmarks covering 44 occupations, compared to 70.9 percent for GPT-5.2. It demonstrates particular strength in spreadsheet modeling, presentation creation, and document editing while reducing hallucinations by 33 percent for individual claims.

Key Features

• Native Computer Use: Operates computers via screenshots, keyboard, and mouse commands or Playwright libraries for autonomous agentic workflows.
• Tool Search Architecture: Lightweight tool lookup system reducing token usage by 47 percent when accessing large tool ecosystems like MCP servers.
• Extended Context Window: Supports up to 1 million tokens for long horizon planning, execution, and verification across complex multi step tasks.
• Reasoning Transparency: Provides upfront thinking plans in ChatGPT, allowing users to adjust direction mid response for better alignment.
• Visual Perception: High resolution image support up to 10.24M pixels with original detail level for precise document parsing and interface interaction.
• Professional Knowledge Work: Specialized capabilities for spreadsheets, presentations, and documents with 87.3 percent accuracy on investment banking analyst tasks.