OpenAI’s CEO Sam Altman admitted that ChatGPT decreased in writing quality to focus on other features with GPT-5.2’s release. However, that didn’t slow the company down from trying to improve ChatGPT in as many areas as possible. On March 5th, 2026, OpenAI announced GPT-5.4, an improved model specifically for professional work.
GPT-5.4 is great at coding, knowledge work, computer use, etc, and it's nice to see how much people are enjoying it.
— Sam Altman (@sama) March 7, 2026
But it's also my favorite model to talk to! We have missed the mark on model personality for awhile, so it feels extra good to be moving in the right direction.
What’s New with GPT-5.4?
GPT-5.4 brings OpenAI’s recent advances in reasoning, coding, and agentic workflows into one convenient model. It builds upon the features of GPT‑5.3‑Codex, specifically focusing on professional tasks, such as those involving spreadsheets, presentations, and documents. The goal of this new version is to get real work completed accurately, effectively, and efficiently with less back and forth between you and the AI.
A few key improvements include:
- Adjusting Course Mid-Response – While GPT‑5.4 Thinking works on answering questions, you can adjust your query to receive more accurate outputs without additional back and forth.
- Improved Deep Web Research – Thinking is now better at maintaining context to help with research, especially for complex questions requiring extra time to process.
- Advanced Computer Use – The AI’s improved capabilities in Codex and API make it easier for agents to operate computers and tackle complex workflows.
- Up to 1M Tokens of Context – GPT-5.4 is the most token-efficient reasoning model, making it easier to plan, execute, and verify tasks across long conversations.
How Does GPT-5.4 Compare to Previous Models?
When compared to past ChatGPT models, GPT-5.4 shows impressive growth, especially in the following categories.
Knowledge Work
GPT-5.4 builds on 5.2’s reasoning capabilities to offer more consistent and polished responses. GPTval tested each version’s abilities to produce accurate knowledge for 44 different occupations. The win rate compared to the industry professional resulted in 83% wins and ties for GPT-5.4 while GPT-5.2 only reached 70.9%.
The new version especially improved when it came to creating and editing spreadsheets, presentations, and documents, which are much more organized and coherent than before.
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT.
— OpenAI (@OpenAI) March 5, 2026
GPT-5.4 is also now available in the API and Codex.
GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model. pic.twitter.com/1hy6xXLAmJ
Computer Use
OpenAI stated that GPT-5.4 is “our first general-purpose model with native computer-use capabilities.” It helps developers build agents that can complete real tasks across different platforms and systems. It can also write code and implement commands.
Previously, GPT-5.2 scored 47.3% accuracy for its ability to navigate a desktop environment, according to OSWorld-Verified. GPT-5.4 reached 75% accuracy, surpassing human performance.
MMMU-Pro tested the model’s visual understanding and reasoning, which further confirmed its strengthened computer use. GPT-5.4 achieved an 81.2% success rate while GPT-5.2 scored 79.5%.
Coding
GPT-5.4’s coding abilities improve greatly in accuracy and estimated latency from GPT-5.2 and slightly from GPT‑5.3‑Codex on SWE-Bench Pro. It’s able to complete coding tasks with less manual intervention. When turning on fast mode, GPT-5.4 achieves 1.5x faster token velocity in Codex.
The model’s coding abilities have been shown on several types of games, including theme park simulations and RPG games. The AI can test the game that it’s building while it’s still building it.
Tool Use
ChatGPT can more easily work with external tools using this new model. Tool search is a new feature that can choose from a list of available tools based on context rather than always needing to be told which tool to use. This requires fewer prompts to get the AI to use effective tools.
GPT-5.4 also improves agentic tool calling, which determines when and how to use tools. The update makes this feature more accurate and requires fewer tool yields than GPT-5.2.
Safety
OpenAI continues to enhance the safety features of ChatGPT with this new version. GPT-5.4 has an expanded cyber safety stack, trusted access controls, and asynchronous blocking for higher-risk requests. You can learn more about GPT-5.4’s cybersecurity capabilities in the system card.
Keep Your Brand Ranking in AI Searches
AI models like ChatGPT frequently receive new updates to improve existing skills and add new abilities. With each update, AI optimization tactics could shift, so keep your marketing strategies current to ensure your brand remains visible in AI searches.
We are the Agency for Influence
Discover new ways to drive revenue and build reputation for your brand.
