OpenAI has introduced GPT-5.4, the latest update in its GPT-5 series of artificial intelligence models. The new release brings improvements in reasoning, coding performance, and automation features. One of the most notable additions is a capability that allows the AI to interact with computer environments and perform complex tasks through automated workflows.
The new models are being made available through ChatGPT and OpenAI’s developer API, allowing both everyday users and developers to benefit from the upgrade.
Also read: OpenAI Hints at GPT-5.4 Release Soon After Launching GPT-5.3 Instant
GPT-5.4 and GPT-5.4 Pro Now Available
OpenAI released two versions of the new model:
- GPT-5.4
- GPT-5.4 Pro
The standard GPT-5.4 model is available in ChatGPT under the name GPT-5.4 Thinking, replacing the previous GPT-5.2 Thinking model for certain subscribers.
Access to the new model currently includes:
- ChatGPT Plus users
- Team subscribers
- Pro subscribers
The more powerful GPT-5.4 Pro version is limited to Pro and Enterprise users, indicating that it is designed for more demanding professional workloads.
Improvements in Reasoning and Coding
The GPT-5.4 release introduces upgrades across several areas of performance. According to OpenAI, the model builds on earlier GPT-5 versions by improving how it handles reasoning tasks and software development.
The model combines elements of previous systems, including the coding abilities introduced in GPT-5.3 Codex, while also expanding its usefulness for general tasks.
This means GPT-5.4 can better assist with:
- Writing and debugging code
- Working with spreadsheets and documents
- Generating presentations or structured reports
- Managing multi-step workflows using tools
These improvements aim to make the model more capable for both everyday productivity and professional development tasks.
New Computer-Use Capability
One of the most significant new features in GPT-5.4 is computer-use functionality. This allows the AI to operate inside a virtual computer environment using automated actions.
Through this capability, the model can:
- Interact with applications
- Perform web browsing tasks
- Control mouse and keyboard actions
- Run commands on a computer system
- Automate complex workflows
Instead of simply providing instructions, the AI can generate code that interacts with a computer interface and executes tasks step by step.
This feature opens the possibility for AI agents that can assist with tasks like data processing, online research, or workflow automation.
Performance Benchmarks
OpenAI shared performance data from internal tests to demonstrate the improvements in GPT-5.4.
On the OSWorld-Verified benchmark, which evaluates how well an AI can navigate a desktop environment, the model achieved a success rate of around 75 percent.
The system also showed strong results in visual reasoning tests. On the MMMU-Pro benchmark, which measures an AI model’s ability to understand visual information, GPT-5.4 reached a score of approximately 81 percent.
These results indicate improvements in how the model interprets images and interacts with graphical computer environments.
Stronger Safety Controls
With the addition of computer-use abilities, OpenAI has implemented additional safety measures to monitor and control the model’s behavior.
The company describes GPT-5.4 as having high cyber capability, meaning extra precautions are necessary when deploying the system.
Safety measures include:
- Monitoring tools for AI actions
- Access controls for trusted environments
- Systems that can block unsafe operations
- New evaluation tools to analyze reasoning transparency
OpenAI has also introduced a framework to test whether the model hides or alters its reasoning process when performing tasks.
Expanding the Role of AI Assistants
The introduction of agent-style computer control represents another step in the evolution of AI systems. Instead of acting only as conversational tools, modern AI models are increasingly designed to perform tasks directly within digital environments.
This shift could allow AI assistants to help with more complex activities, such as automating repetitive work, managing software tools, or coordinating multi-step tasks across applications.
Also read: Google Launches Gemini 3.1 Flash-Lite, Its Fastest and Most Affordable AI Model Yet
Final Thoughts
The release of GPT-5.4 and GPT-5.4 Pro marks another major milestone in OpenAI’s development of advanced AI systems. With improvements in reasoning, coding, and computer interaction, the new models aim to support more sophisticated tasks for both individuals and businesses.
Although some features are currently limited to certain subscription tiers, GPT-5.4 highlights the growing trend of AI systems moving beyond conversation into real digital task automation.