OpenAI Launches GPT-5.4: Most Advanced Model Yet for Professional Tasks

On Thursday, March 5, 2026, OpenAI introduced GPT-5.4 Thinking, which it claims to be its most sophisticated model for professional tasks to date. In ChatGPT, Plus, Team, and Pro subscribers now have access to GPT-5.4 Thinking, the reasoning model that replaces GPT-5.2 Thinking. This model coexists with GPT-5.3 Instant, which was launched two days earlier for conversational interactions, with an automatic mode handling the transition between the two. OpenAI also offers GPT-5.4 Pro for more complex tasks (available only to Pro and Enterprise plans) and a standard version of GPT-5.4 in its API and Codex.

Innovations in Computing, Coding, and Tool Management with GPT-5.4

GPT-5.4: A Generalist Model Tailored for Agents

GPT-5.4 integrates the latest advancements from OpenAI in reasoning, coding (incorporating capabilities from GPT-5.3-Codex), and tool management into a single model. The company describes 5.4 as a model that “executes complex tasks precisely and efficiently, delivering what is requested with fewer interactions”. Notably, this is OpenAI’s first general-purpose model that natively includes computer use capabilities, meaning it can control a computer by executing clicks, keyboard commands, and interpreting screenshots, as demonstrated in the video below.

Additional notable enhancements include:

• A new tool-search system in the API, which allows the model to dynamically search for tool definitions instead of loading them all at once, significantly reducing token consumption.
• An expanded context window that can accommodate up to a million tokens in Codex and the API.
• Improved token efficiency, with GPT-5.4 expected to solve the same problems using fewer tokens than GPT-5.2, leading to faster responses and reduced costs.

In ChatGPT, the Thinking version now displays a preliminary work plan for complex queries, with the option to adjust instructions during the response. OpenAI’s goal is to “more easily guide the model to the exact desired outcome without starting from scratch or requiring additional interactions”. The company also claims a reduction in hallucinations compared to GPT-5.2, with responses reportedly less likely to contain factual errors, strongly supported by benchmarks.

ChatGPT for Excel, Another Announcement of the Day

Simultaneously with the launch of GPT-5.4, OpenAI has also introduced ChatGPT for Excel in beta, an add-on that integrates ChatGPT directly into spreadsheets, similar to what Claude already offers. This tool allows users to build, analyze, and update models in natural language while maintaining the structure and formulas of the workbook. Currently, this feature powered by GPT-5.4 is available only to Business, Enterprise, Edu, Pro, and Plus subscribers in the United States, Canada, and Australia. A version for Google Sheets has been announced.

OpenAI Continues to Launch New Products Amid Increased Competition

This launch comes just two days after the release of GPT-5.3 Instant, an update aimed at addressing the perceived condescending tone of ChatGPT by many users. This acceleration in release cadence is not insignificant.

For several months, OpenAI has been experiencing a decline in its dominant position. A survey we conducted in February showed that 54% of respondents noticed a decline in the quality of ChatGPT, and 64% of those who were disappointed had already switched to an alternative. Meanwhile, Anthropic is capitalizing on this trend: Claude has recently topped the American App Store downloads and introduced Import Memory, a tool that facilitates migration from ChatGPT.

GPT-5.4 is OpenAI’s technical response to this competitive pressure. However, the claimed improvements by the company rely largely on its own internal assessments, an indicator that should be taken with caution as standardized evaluations of AI models are currently facing a credibility crisis.