OpenAI is releasing a new model today, and like GPT-5.2 before it, GPT-5.4 is all about professional work. OpenAI is calling GPT-5.4 its most capable frontier model for tasks like coding and data analysis. OpenAI claims the new model produced presentations with stronger, more varied aesthetics and made more effective use of its image generation tools.
It’s also the first model from OpenAI built with native computer-use capabilities, making it better at carrying out tasks across several apps at the same time. When it comes to computer use, one noticeable improvement OpenAI has recorded is the way GPT-5.4 issues mouse and keyboard commands. It’s significantly better at navigating a desktop environment than its predecessor.
When users turn to GPT-5.4 in ChatGPT, where it will now be the default model for the chatbot’s Thinking mode, the system will outline how it plans to tackle a request, giving people the opportunity to tell it to adjust course as it’s generating a response. At the same time, the new model offers better web research capabilities, especially when it comes to “highly specific” queries, according to OpenAI.
“Together, these improvements mean higher-quality answers that arrive faster and stay relevant to the task at hand,” the company states. Separately, OpenAI claims GPT-5.4 is its most factual model yet, noting, relative to GPT-5.2, it’s 18 percent less likely to generate a response with any errors. Here’s hoping it knows not to turn to Grokipedia for information, something its predecessor was known to do.
As mentioned, GPT-5.4 will be available in ChatGPT when users select the chatbot’s Thinking mode, and as GPT-5.4 Pro from the model picker. As such, this isn’t a release for Free and Go users — or even Plus subscribers, for that matter. It’s more for enterprise customers, and developers who rely on the company’s Codex app. On that note, for API customers, OpenAI claims GPT-5.4 is its most token efficient reasoning model to date, though those tokens will cost more than their GPT-5.2 counterparts. For instance, OpenAI is pricing one million input tokens at $2.50, up from $1.75 with GPT-5.2.
The fact that OpenAI has increasingly shifted its strategy to focus on professionals shouldn’t be surprising. When Microsoft announced last September it would add Anthropic’s models to Copilot 365 (where previously it depended exclusively on OpenAI’s systems), there were reports that suggested the company made the decision because it found Claude was better at tasks like generating spreadsheets and presentations. Reporting from The Information suggests OpenAI is generating about $25 billion in annualized revenue. However, the company is still far from being profitable, and with more than $1.4 trillion in data center commitments on the books, it’s reliant on funding from investors to keep the lights on. Seen in that context, productivity represents a place where it might have a chance to build a sustainable business.
