GPT-5.4 Release: Advanced AI Model with Computer Use and Professional Work Capabilities
Key Points
- First AI model with native computer-use capabilities
- 83% success rate matching industry professionals
- 1M token context window for complex workflows
Summary
GPT-5.4 is OpenAI's latest frontier model designed for professional work, combining advanced reasoning, coding, and agentic workflows. Available in ChatGPT (as GPT-5.4 Thinking), API, and Codex, it represents a significant leap in AI capabilities for real-world applications.
Key Features
- Native Computer Use: First general-purpose model with state-of-the-art computer-use capabilities, enabling agents to operate computers and execute complex workflows
- Extended Context: Supports up to 1M tokens of context for long-horizon task planning and execution
- Enhanced Professional Work: Improved performance on spreadsheets, presentations, and documents with 87.3% mean score on investment banking analyst tasks
- Token Efficiency: Most token-efficient reasoning model yet, using significantly fewer tokens than GPT-5.2
- Improved Visual Perception: Enhanced screenshot interpretation and document parsing with support for up to 10.24M pixel images
Performance Benchmarks
- GDPval: 83.0% wins/ties (vs 70.9% for GPT-5.2) across 44 professional occupations
- OSWorld-Verified: 75.0% success rate on desktop navigation tasks, exceeding human performance (72.4%)
- SWE-Bench Pro: 57.7% on software engineering tasks
- Factual Accuracy: 33% fewer false claims and 18% fewer response errors compared to GPT-5.2
Developer Access
- Available through ChatGPT, API, and Codex
- New computer tool in API with customizable safety policies
- Priority processing for faster speeds
- Updated spreadsheet and presentation skills
- Experimental Playwright Interactive skill for visual debugging