OpenAI has released GPT-5.4, its newest frontier artificial intelligence model designed for professional tasks across coding, research, and document work. The company described the system as its most capable and efficient model for real world workloads. The release arrives simultaneously across ChatGPT, the company’s developer API, and the Codex platform, marking one of OpenAI’s largest model launches this year.
The rollout follows growing scrutiny of OpenAI after its decision to work with the United States Department of Defense. The company appears focused on reinforcing its leadership in artificial intelligence development through a major product release.
OpenAI launched GPT-5.4 in three versions designed for different workloads. The standard model targets general tasks inside ChatGPT and developer tools. GPT-5.4 Thinking focuses on complex reasoning problems and replaces GPT-5.2 Thinking for ChatGPT Plus, Team, and Pro subscribers. GPT-5.4 Pro targets enterprise users who require maximum performance on demanding workloads.
The API version supports context windows of up to one million tokens. This capacity allows the model to analyze extremely large datasets, documents, and code bases in a single request.
GPT-5.4 also introduces native computer use capabilities. The model can operate software across a computer through screenshots, keyboard inputs, and mouse actions. These abilities allow AI agents to navigate websites, manage files, and complete multi step workflows across different applications without constant human supervision.
“This makes it easier to guide the model toward the exact outcome you want without starting over or requiring multiple additional turns,” OpenAI says.
OpenAI combined several previous advances inside the new model. GPT-5.4 integrates strong coding performance from earlier Codex systems while improving how the model interacts with software tools, spreadsheets, presentations, and documents used in professional environments.
Benchmark results show significant improvements. GPT-5.4 reached a 75% score on a computer use evaluation compared with 47.3% for GPT-5.2. The system also improved performance on web search tasks and professional knowledge work tests.
OpenAI said the model produces responses with fewer mistakes. Individual responses show a 33% reduction in errors compared with GPT-5.2, while overall error rates fall by roughly 18%.
The company also introduced a new tool search system. Instead of loading every available tool at once, the model retrieves tools only when needed. Internal testing across hundreds of tasks reduced total token usage by 47% while maintaining accuracy.
The release intensifies competition in enterprise artificial intelligence software as companies race to deploy systems capable of performing real professional work across industries. Meanwhile, just days ago, OpenAI got under massive public critique for going with a Pentagon deal that essentially allows US government to use their AI for mass surveillance and warfare.
