OpenAI has launched its most advanced iteration of ChatGPT yet, introducing GPT-5.4 Thinking and GPT-5.4 Pro. The release signals a continued push to enhance AI capabilities for professional use, directly responding to competition from Google and Anthropic. The fresh models promise significant improvements in coding, reasoning and the ability to interact with software, marking a substantial leap forward for the popular chatbot.
The rollout of GPT-5.4 follows the release of GPT-5.3 Instant earlier this month, designed for faster responses. Prior to that, ChatGPT had remained on version 5.2 for several weeks, impacting the experience for its over 900 million users. OpenAI’s rapid succession of updates underscores the intense competition in the AI landscape.
GPT-5.4 builds upon previous advancements, merging the coding prowess of GPT-5.3 Codex with the reasoning abilities of OpenAI’s “Thinking” models. This combination unlocks new functionalities, including a unique interface feature in GPT-5.4 Thinking that displays a detailed plan of its thought process before responding. Users can then refine this plan mid-response, steering the AI towards a more accurate or relevant outcome without restarting the prompt – a step OpenAI describes as furthering the development of agentic AI.
OpenAI is touting benchmark results that demonstrate GPT-5.4’s improved performance. On the GDPval test, which measures performance on professional tasks like finance, law, and marketing, GPT-5.4 achieved an 83% success rate, a significant increase from the 70.9% achieved by version 5.2. The company also claims a substantial reduction in “hallucinations” – instances where the AI generates incorrect or nonsensical information – with results comparable to or exceeding those of Anthropic and Google’s models.
A Million Tokens and Computer Control
A key upgrade with GPT-5.4 is its expanded context window, now supporting up to 1 million tokens. This allows the model to analyze entire codebases, lengthy document collections, or complex agent workflows within a single request, according to OpenAI’s announcement. GPT-5.4 is the first “mainline model” with built-in computer use capabilities, enabling agents to directly interact with software to complete, verify, and fix tasks in a build-run-verify-fix loop.
This “computer control” feature allows ChatGPT to autonomously navigate the web, populate Excel spreadsheets, and create PowerPoint presentations based on screen captures. This functionality is seen as a direct response to features offered by Anthropic’s Claude, intensifying the competition in the AI assistant market.
GPT-5.4 is launching, available now in the API and Codex and rolling out over the course of the day in ChatGPT.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.
— Sam Altman (@sama) March 5, 2026
For users requiring maximum performance, OpenAI is also offering GPT-5.4 Pro, a flagship model optimized for complex mathematical and scientific calculations. This version is available to subscribers of ChatGPT Pro and Enterprise plans.
New Integrations for Financial Professionals
The release of GPT-5.4 also includes specialized capabilities for the financial industry. The model now integrates directly with FactSet Research Systems and Third Bridge, providing professionals with access to comprehensive data for financial analysis and investment documentation, as reported by Parameter.io. OpenAI has also launched a dedicated ChatGPT add-in for Microsoft Excel, further streamlining workflows for financial professionals.
According to OpenAI, GPT-5.4 achieved an 87.3% score on an internal investment banking spreadsheet assessment, a significant improvement over GPT-5.2’s 68.4% result. The model also demonstrated leading performance on Mercor’s APEX-Agents benchmark, designed to test professional skills in law and finance, according to a statement from Mercor CEO Brendan Foody. “[GPT-5.4] excels at creating long-horizon deliverables such as slide decks, financial models, and legal analysis,” Foody said, “delivering top performance although running faster and at a lower cost than competitive frontier models.”
The deployment of GPT-5.4 is currently rolling out and is expected to be completed in the coming days. GPT-5.4 Thinking is available to ChatGPT Plus, Team, and Pro subscribers, replacing GPT-5.2 Thinking, which will be phased out in three months.
As OpenAI continues to refine its AI models, the focus remains on enhancing performance, reducing errors, and expanding the capabilities of ChatGPT to meet the evolving needs of professionals across various industries. The ongoing competition with rivals like Google and Anthropic is driving rapid innovation in the field, promising even more powerful and versatile AI tools in the future.
What impact will these new capabilities have on productivity and workflows in the coming months? Share your thoughts in the comments below.