Jakarta, INTI - OpenAI has introduced its latest artificial intelligence (AI) model, GPT-5.4, on Thursday, March 5, 2026, U.S. time.
In an announcement published on its official website, the company stated that the new model was designed to handle a wide range of professional tasks more efficiently.
“Today we are releasing GPT-5.4 (as GPT-5.4 Thinking), the API, and Codex. This is our most capable and efficient frontier model for professional work,” OpenAI wrote in the announcement.
GPT-5.4 combines improved reasoning capabilities, advanced coding performance, and the ability to assist with various professional tasks involving spreadsheets, documents, and presentations.
One of the key updates introduced with the model is a new capability that allows AI to directly operate a computer.
With this feature, GPT-5.4 can perform tasks on behalf of users by interacting with multiple applications and completing workflows automatically.
“This model can complete complex real-world tasks accurately, effectively, and efficiently, allowing users to achieve results with minimal switching between applications,” OpenAI added.
The system is also capable of generating code to execute commands through the keyboard and mouse, using information obtained from screenshots.
In addition, the model is said to have improved capabilities in using web browsers and calling various tools and APIs more accurately to complete tasks.
OpenAI also integrated GPT-5.4 into its API and its AI-powered coding tool, Codex, further expanding the model’s functionality for developers.
The ability to issue and execute commands represents a significant advancement in AI model capabilities.
GPT-5.4 can now utilize tools and API access more precisely and efficiently when completing different types of tasks.
The model also reflects a broader shift toward agent-based computing, where AI agents are able to carry out increasingly complex workflows autonomously.
This means AI is no longer limited to performing simple tasks, but can increasingly operate in the background of a computer to execute more complex processes across multiple software applications.
More Accurate Responses
GPT-5.4 is also claimed to deliver more accurate answers by incorporating a broader process of gathering information from various sources across the internet.
OpenAI stated that its latest model is “more persistent” in searching for data to determine which sources are most relevant when responding to complex questions.
This capability makes the model particularly suitable for addressing complicated queries and transforming analytical findings into comprehensive and well-structured explanations.
With these improvements, OpenAI claims GPT-5.4 is its most factual AI model to date, reportedly reducing errors by about 33 percent compared with GPT-5.2.
“Thinking” Mode Introduced
The new model will appear as an option called “GPT-5.4 Thinking” when users access the ChatGPT chatbot. This version is positioned as an AI agent specialized in handling complex reasoning tasks.
It is designed to answer more advanced queries and allows users to modify or refine their requests while the response process is still ongoing.
“This will make it easier for users to guide the model toward the desired outcome without having to start over or repeatedly perform additional reasoning,” OpenAI explained.
GPT-5.4 is already available on the web version of ChatGPT and on Android mobile devices, while availability for iOS will follow at a later time.
The GPT-5.4 Thinking model is currently accessible to users subscribed to the Plus, Team, and Pro plans.
Meanwhile, GPT-5.4 Pro, offering maximum performance for handling complex tasks, has been released for ChatGPT Enterprise and Edu users.
Conclusion
The launch of GPT-5.4 marks another step forward in the evolution of AI-driven computing. With stronger reasoning capabilities, improved accuracy, and the ability to interact directly with computer systems, the model reflects a shift toward agent-based AI that can handle complex tasks autonomously. These advancements signal how artificial intelligence is gradually moving beyond simple question-answering tools to become an active digital assistant capable of supporting professional workflows across multiple applications.
Read more: United States Prepares New Regulations to Tighten Overseas Exports of AI Chip