US-based AI giant OpenAI has rolled out the GPT-5.4. The company claims that it is the most capable and efficient frontier model so far. The company further said that the new system combines advances in reasoning, coding, and agentic workflows into one model designed to handle complex professional tasks with fewer errors and less back and forth.
GPT- 5.4 Versions
The new update is available in two version that is Thinking and Pro. The Thinking version introduces a new way of interacting with the model by showing an upfront plan of its reasoning by which users can adjust the direction mid-response. The company says that it makes the result more relevant and reduces the need for multiple turns. It also enhances deep web research, particularly for highly specific queries along with context over longer conversation.
The Pro version is developed for professionals who need maximum performance on complex tasks.
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT.
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
— OpenAI (@OpenAI) March 5, 2026
GPT 5.4 Integration with API and Codex
Apart from ChatGPT, the GPT-5.4 is also integrated into the API and Codex. The company claims that this is the first general-purpose model the company has launched with native computer use capabilities which allow agents to use software directly and carry out workflows across applications. The new update supports up to 1 million tokens of context, allowing long-horizon planning and execution. OpenAI also highlights new tool search features, which help agents find and use the right connectors more efficiently.
GPT 5.4 Performance
The performance benchmark portrays GPT-5.4 faster than the past models. On GDPval the GPT scored 83 per cent wins or ties, compared to 70.9 per cent for GPT-5.3 codex and GPT-5.2.
On OSWorld-Verified, the latest GPT has reached 75 per cent, which is a significant jump from GPT-5.2’s 47.3 per cent. Similar kind of growth is also observed across SWE-Bench Pro, Toolathlon, and BrowseComp
The company has positioned the latest GPT as a step forward, making AI more reliable for professional work. If we compare the latest update with GPT 5.3 which is mainly focused on coding and GPT-5.2 which is less efficient with tokens, the new GPT-5.4 combines stronger reasoning, broader professional support, and agentic computer-use capabilities. The AI giant claims that this translates into faster developer workflows, more reliable agents, and higher-quality output across ChatGPT, the API, and Codex.
Syed Ziyauddin is a media and international relations enthusiast with a strong academic and professional foundation. He holds a Bachelor’s degree in Mass Media from Jamia Millia Islamia and a Master’s in International Relations (West Asia) from the same institution.
He has work with organizations like ANN Media, TV9 Bharatvarsh, NDTV and Centre for Discourse, Fusion, and Analysis (CDFA) his core interest includes Tech, Auto and global affairs.
Tweets @ZiyaIbnHameed
