OpenAI has introduced GPT-5.4, a new artificial intelligence model the company describes as its most advanced and thoughtful system to date. In a press release, OpenAI said the model represents its “most capable and efficient frontier model for professional work,” signaling a major step forward in how AI can assist businesses and knowledge workers.
One of the most notable new capabilities in GPT-5.4 is the ability for users to interrupt the model while it is generating a response. If the system begins heading in the wrong direction or misinterprets a prompt, users can stop it mid-response and redirect it. According to OpenAI, this feature is particularly useful for complex tasks that require multiple steps or detailed reasoning.
When used within ChatGPT, GPT-5.4 automatically activates a “thinking mode,” which relies on an internal reasoning process to determine the best way to answer a prompt. As part of this system, the model generates a short “preamble” before completing more complicated requests. This preamble gives users a chance to review and interrupt the process early if the AI begins researching or analyzing topics that are only loosely connected to the original question.
OpenAI says the feature is designed to address a common issue with AI tools: occasionally drifting into unrelated topics when handling complex instructions. By allowing users to intervene early, the company believes GPT-5.4 will be easier to guide and more reliable during long, multi-step workflows.
The new model also comes with improvements to ChatGPT’s deep research feature, which compiles extensive reports by analyzing large volumes of online information. With GPT-5.4, OpenAI says the system is better able to maintain the context of the original prompt even when processing hundreds of documents over extended periods of time. This enhancement is intended to make AI-generated research more coherent and accurate.
Industry experts have already highlighted the model’s potential for professional use cases. Niko Grupen, head of applied research at legal AI startup Harvey, said in a statement that “GPT-5.4 sets a new bar for document-heavy legal work.” He added that the system is better at structuring complex transactional analysis, maintaining accuracy across lengthy contracts, and producing the level of detail required by legal professionals.
For companies that integrate OpenAI models into their products through the API, GPT-5.4 also offers greater efficiency. The company says the model uses significantly fewer tokens to complete tasks compared with its predecessor, GPT-5.2. Tokens represent the units of data that AI models process and generate, and they directly influence the cost of using an AI system. By reducing token usage, businesses can lower operational costs while maintaining performance.
Alongside the model launch, OpenAI announced a suite of new enterprise tools designed to streamline data analysis and decision-making. According to the company, these tools aim to reduce manual analysis, simplify workflows, and accelerate the process of turning data into actionable insights.
Among the new offerings are custom versions of ChatGPT integrated into Excel and Google Sheets. These AI-powered tools can build, analyze, and edit spreadsheets, helping professionals automate tasks that traditionally require significant manual effort. ChatGPT can also follow structured workflows to perform tasks such as creating earnings reports, analyzing cash flow, and drafting investment memos.
OpenAI also says GPT-5.4 delivers state-of-the-art computer-use capabilities, enabling the AI to complete complex agent-like workflows that span multiple websites, data sources, and applications. In testing on GDPval, an OpenAI benchmark that evaluates how well AI agents perform knowledge-work tasks, GPT-5.4 matched or exceeded the output of industry professionals in 83 percent of comparisons. That marks a significant improvement over GPT-5.2, which achieved a score of 71 percent.
The company claims the new model also outperformed Claude Opus 4.6, the flagship AI model from Anthropic, on several benchmarks measuring capabilities in computer use, science, and mathematics.
GPT-5.4 is now available through ChatGPT, OpenAI’s API, and Codex, the company’s platform for managing AI-powered coding agents. OpenAI has also released GPT-5.4-Pro, a larger and more powerful version of the model available on ChatGPT and through the API.
The new model comes with higher pricing compared with its predecessor. GPT-5.4 costs $2.50 per million input tokens and $15 per million output tokens, up from GPT-5.2’s pricing of $1.75 and $14 respectively. The premium GPT-5.4-Pro model is significantly more expensive, priced at $30 per million input tokens and $180 per million output tokens.
Despite the higher cost, OpenAI says the combination of improved efficiency, advanced reasoning, and expanded workflow capabilities positions GPT-5.4 as a powerful tool for professionals who rely on AI for research, analysis, and complex decision-making.