OpenAI brings GPT-5.4 for complex workflows
GPT-5.4 is OpenAI’s most capable model for professional tasks, featuring native computer-use capabilities and advanced agentic workflows to complete complex, multi-step goals accurately.
OpenAI has introduced a potential solution to modern workloads with the launch of GPT-5.4, a new frontier model available in ChatGPT, the API, and Codex. It is designed specifically to handle the complexities of professional work.
Unlike its predecessors, GPT-5.4 is built to act as an agent, which is an AI that can complete multi-step goals with far less human help, allowing it to operate software and solve problems directly.
The release is a step forward from GPT-5.2 and other earlier versions. While GPT-5.2 was a powerful reasoning tool, GPT-5.4 combines those abilities with the advanced coding skills found in GPT-5.3-Codex, making it a more versatile tool for employees who need to manage spreadsheets, presentations, and documents simultaneously.
In industry tests, the model matched or exceeded the performance of human professionals in 83% of comparisons, which is a notable jump from the 70.9% achieved by previous models.
“It’s much better at knowledge work and web search, and it has native computer use capabilities. You can steer it mid-response, and it supports 1m tokens of context,” remarked Sam Altman, OpenAI CEO.
One of the new features is native computer-use capability, allowing the AI to perceive a computer screen and interact with it by moving the mouse or typing on the keyboard. It means the model can now navigate websites and use software systems just as a person would.
On the OSWorld-Verified benchmark, which tests how well an AI can handle a desktop environment, GPT-5.4 achieved a 75% success rate. This performance actually beats the human average of 72.4% in the same test.
OpenAI highlighted that the impact of these improvements is already being felt by early users in various industries.
In addition to computer use, the model introduces a feature called tool search to help it manage vast ecosystems of software connectors. In the past, AI models had to be given every tool definition at the start of a chat, which was like trying to carry a whole toolbox when you only need a single screwdriver.
With tool search, GPT-5.4 can look up the specific instructions for a tool only when the moment is right. This reduces the number of tokens, which are the small units of text or data the AI processes, and makes tasks faster and cheaper.
The model also features an expanded context window, which is essentially the amount of information the AI can hold in its active memory at once. In Codex, developers can now experiment with a context window of one million tokens. This allows for the analysis of massive sets of data or very long documents without the AI losing track of earlier points.
The coding community has also seen a boost in performance. Through a new fast mode in Codex, the model can generate code up to 1.5 times faster while maintaining its high level of intelligence. Lee Robinson, the VP of Developer Education at Cursor, highlighted the model's assertive nature.
“GPT-5.4 is currently the leader on our internal benchmarks. Our engineers find it to be more natural and assertive than previous models. It works through ambiguous problems without second-guessing itself, and it's proactive about parallelizing work to keep things moving,” he noted.
It is also the most factual model OpenAI has released, so it is less likely to produce false information or hallucinations than GPT-5.2.
OpenAI has also made the model more steerable, meaning it is easier to guide toward a specific result.
In ChatGPT, the AI now provides a plan of its thinking before it starts a long task, allowing users to adjust its direction in the middle of a response. This prevents the need to start a conversation over if the AI drifts off course.
While GPT-5.4 is priced higher per token than GPT-5.2, its increased efficiency can result in lower total costs for complex jobs. The model is currently rolling out to paid ChatGPT users and developers via the API.
OpenAI is also releasing GPT‑5.4 Pro in ChatGPT and the API, for people who want maximum performance on complex tasks.
Recently, Perplexity AI launched Computer, describing it as a fully autonomous digital worker that plans, delegates and executes multi-step projects end to end.
GPT-5.4 launch comes a week after the ChatGPT maker secured a staggering $110 billion in new investment at a pre-money valuation of $730 billion. The funding round was led by major strategic partners including Amazon, which is investing $50 billion, alongside SoftBank and NVIDIA, who are each contributing $30 billion.