GPT-5.4 Pro Officially Drops March 5 – Pro Tasks Supercharged!
The new model achieves a state-of-the-art 83% on real-world job tasks and can control software via screenshots.
OpenAI has officially launched GPT-5.4 Pro, a model explicitly engineered for professional-grade productivity. In internal benchmarks using the GDPval test of real-world job tasks, it achieved a state-of-the-art 83.0% success rate, a significant 12.1-point leap over GPT-5.2's 70.9%. The model excels at generating complex documents, legal analyses, and business materials with fewer errors, scoring 91% on a specific legal-document benchmark. A groundbreaking feature is its 'native computer-use' capability, allowing AI agents powered by GPT-5.4 to interpret screenshots and issue mouse/keyboard commands to autonomously browse websites, fill forms, and manipulate software. This marks a major step toward practical automation.
Beyond raw power, GPT-5.4 is designed for efficiency, using significantly fewer tokens than its predecessor for the same tasks, translating to faster and more cost-effective API calls. OpenAI is embedding these capabilities directly into user tools, releasing a dedicated ChatGPT-for-Excel add-in to bring advanced AI assistance into analysts' spreadsheets. Available through the API as `gpt-5.4` and `gpt-5.4-pro` and in ChatGPT, this release signals a shift from experimental AI to executable, integrated business solutions, setting a new bar for AI-driven professional work.
- Achieved 83.0% success rate on GDPval professional task benchmark, up from 70.9% for GPT-5.2
- Features 'native computer-use' to control software UIs via screenshots for autonomous task automation
- More token-efficient than prior models, leading to faster performance and lower API costs
Why It Matters
It transforms AI from a conversational tool into an autonomous agent capable of executing complex, multi-step professional workflows.