OpenAI released GPT-5.4 on March 5, 2026 as a frontier model for professional work across ChatGPT, the API, and Codex. The update matters because it combines stronger reasoning, coding, computer use, tool workflows, spreadsheet/document/presentation work, and long-context API use in one mainline model.
Quick answer: GPT-5.4 is best for complex professional workflows: coding agents, computer-use automation, deep research, spreadsheets, presentations, document-heavy tasks, and long-horizon tool use. Casual users may not need it; teams doing real work across files, tools, and code will benefit most.
| GPT-5.4 area | What changed | Who cares most |
|---|---|---|
| ChatGPT | GPT-5.4 Thinking for harder work | Power users, analysts, operators |
| API | gpt-5.4 and gpt-5.4-pro |
Developers and product teams |
| Codex | Stronger coding and agent workflows | Engineers and vibe-coding teams |
| Computer use | Native desktop/browser interaction | Agent builders and automation teams |
| Context | Around 1M-token class context in API docs | Long-document and enterprise workflows |
| Professional output | Better spreadsheets, docs, presentations | Finance, ops, consulting, legal, GTM |
Source check — June 4, 2026: OpenAI’s official Introducing GPT-5.4 post says GPT-5.4 rolled out to ChatGPT, the API, and Codex. The official GPT-5.4 API model page lists current context and pricing details. Use those pages for live availability and cost before building production workflows.
For comparisons, see best AI chatbots, Claude Opus 4.8, Claude Code vs Codex, and best AI agents for coding.
What’s New in GPT-5.4?
Onboard computer functionalities
GPT-5.4 has become OpenAI’s first general-purpose model with built-in computer-use capabilities. The new model can interact with websites, work with applications, issue mouse and keyboard commands in response to screenshots, and write code to automate tasks. In navigation tests (OSWorld-Verified), the model achieved a 75% success rate, outperforming the previous version and even exceeding the average human result of 72.4%.
“Its behavior is steerable via developer messages, meaning that developers can adjust behavior to suit particular use cases. Developers can even configure the model’s safety behavior to suit different levels of risk tolerance by specifying custom confirmation policies,” – OpenAI emphasized.
Even higher-quality work on professional tasks
Developers improved the model’s ability to create presentations, edit spreadsheets, and work with documents. In OpenAI’s GDPval test, which evaluates AI agents across 44 different professions, GPT-5.4 set a new record. The model matched or exceeded the performance of industry specialists in 83% of cases, compared to 70.9% for GPT-5.2. Additionally, in test assignments evaluating presentations, 68% of experts preferred GPT-5.4 over GPT-5.2, citing its more attractive visual design, use of diverse graphic elements, and high-quality image generation.
Developer improvements
GPT-5.4 handles complex development tasks better thanks to enhanced reasoning and computer-use capabilities. In SWE-Bench Pro tests, the model performed at levels comparable to GPT-5.3 Codex, while working faster. According to OpenAI, GPT-5.4 excels at frontend development, producing application and website interfaces that look better than with previous models. Another important update is the /fast mode in Codex, which accelerates code generation by 1.5x.
What’s new in tools
Tools are external services and programs that the model can access while completing a task. Previously, the model received descriptions of all available tools upfront, which slowed down processing and increased the request size. With the introduction of Tool Search in GPT‑5.4, the model now receives only a compact list of tools and can load detailed information about a specific tool only when it needs to use it.
Moreover, the new model uses tools more intelligently, especially for multi-step workflows involving multiple services. GPT-5.4 demonstrates higher accuracy and completes tasks in fewer steps compared to previous versions.
Taking information processing to a new level
OpenAI reports that GPT-5.4 improves deep web search capabilities, including for complex and specialized queries. The model can conduct multi-stage searches to provide the most relevant results and can gather information from multiple sources and synthesize it into a single, coherent answer. Moreover, the new model allows query adjustments. GPT-5.4 can present its plan of action for complex tasks, and users can modify the response mid-generation without starting over or repeatedly adding clarifications.
How to start using GPT-5.4?
The new model is already available to all users. Developers can also access it via the API under the name “gpt-5.4”, and use GPT‑5.4 Pro through the API as “gpt‑5.4-pro” for the most complex and resource-intensive tasks.
In ChatGPT, users on Plus, Team, and Pro plans now have access to GPT‑5.4 Thinking, which replaces GPT‑5.2 Thinking. The older GPT‑5.2 Thinking will remain in the Legacy Models section for paid users for another three months, after which it will be fully retired on June 5, 2026.
What are ChatGPT models anyway and why should you care?
It’s easy to get lost among the growing number of AI chatbots and tools released almost daily. While many of us use ChatGPT regularly, far fewer understand how it is structured – and how to get the most out of it with minimal effort.
Since launching ChatGPT, OpenAI has released around ten major model versions along with multiple optimized configurations. It is important to understand one key distinction: ChatGPT delivers the chat experience, supported by AI models that drive the reasoning, speed, and accuracy of its answers. To see how GPT-5.4 compares to other leading models, you can review comparisons like Claude vs ChatGPT or Grok vs ChatGPT.
The specific model connected to the chat determines how responses are generated – whether they prioritize deeper reasoning, faster replies, lower hallucination rates, or improved tone and conversational flow. Each new model version represents an enhancement in one or several of these areas. For example, GPT-5.2 Instant was optimized for speed and efficiency in everyday tasks, while GPT-5.2 Thinking focused on solving complex problems more effectively through extended reasoning.
In this context, GPT-5.4 is not simply another routine update. It reflects a more targeted refinement of user experience. Understanding which model powers your interactions allows you to use AI more strategically rather than passively.