![GPT54]()
OpenAI has unveiled GPT-5.4, the latest addition to its GPT-5 series, positioning it as the company’s most capable and efficient frontier model for professional knowledge work. The model is rolling out across ChatGPT, the OpenAI API, and Codex, with an enhanced GPT-5.4 Pro version offering maximum performance for complex tasks.
The release continues OpenAI’s push toward AI systems that can assist with real-world workflows — from document creation and spreadsheet analysis to advanced research and automation tasks.
Built for Real-World Professional Tasks
According to OpenAI, GPT-5.4 is designed specifically for knowledge workers, focusing on tasks professionals perform daily such as preparing presentations, analyzing financial data, and drafting detailed reports.
On the GDPval benchmark, which evaluates AI performance across tasks representing 44 different occupations, GPT-5.4 matched or exceeded human professionals in 83% of comparisons, up from 70.9% for GPT-5.2.
The benchmark includes real work products like:
These results suggest OpenAI is focusing less on theoretical benchmarks and more on practical productivity tasks.
Native Computer-Use Capabilities
A major upgrade in GPT-5.4 is native computer-use capability, enabling the model to interact with computers and execute multi-step workflows across different applications.
This allows AI agents powered by GPT-5.4 to:
Navigate desktop environments
Use keyboard and mouse actions
Interact with software tools
Execute long-running workflows automatically
The model can also process up to one million tokens of context, enabling it to analyze long documents or manage extended tasks across multiple steps.
In the OSWorld-Verified benchmark, which evaluates AI’s ability to control a desktop environment, GPT-5.4 achieved a 75% success rate, surpassing both previous versions and even human performance in that benchmark.
Stronger Performance in Documents, Presentations, and Spreadsheets
OpenAI emphasized improvements in productivity-focused tasks such as creating spreadsheets, presentations, and documents.
For example:
In spreadsheet modeling tasks similar to those performed by junior investment banking analysts, GPT-5.4 scored 87.3%, compared to 68.4% for GPT-5.2.
In presentation generation tests, human reviewers preferred GPT-5.4 outputs 68% of the time due to better structure and visuals.
These improvements are aimed at making AI a more reliable tool for analysts, consultants, legal professionals, and other knowledge workers.
Fewer Errors and Improved Factual Accuracy
OpenAI says GPT-5.4 also delivers improvements in accuracy and reliability.
On internal evaluations of user-flagged prompts:
Reducing hallucinations remains a major focus for AI developers as these systems increasingly assist with research, business analysis, and technical documentation.
Multi-Step Tool Use and Agent Workflows
The model is also optimized for multi-step tool use, an emerging trend in AI where systems combine reasoning with external tools and APIs.
According to OpenAI partners testing the model, GPT-5.4 demonstrates stronger persistence in completing complex workflows — finishing tasks that earlier models often abandoned midway.
This capability is particularly important for AI agents that must:
Retrieve information from multiple sources
Run calculations or simulations
Generate structured outputs like dashboards or reports
A New Phase in AI Productivity
GPT-5.4 represents OpenAI’s latest attempt to move beyond conversational AI toward autonomous digital workers capable of executing professional tasks end-to-end.
By combining advanced reasoning, long context windows, computer interaction, and improved factual reliability, the company is positioning GPT-5.4 as a platform for building the next generation of AI-powered productivity tools.
✅ In short: GPT-5.4 is less about chat and more about getting real work done — from financial modeling to legal analysis and automated workflows.
Source: OpenAI