{"id":"f-73cd9b4b27-ca940bbf-efde-579d-a56b-d65e66013c74","title":"OpenAI launches GPT-5.4 AI model days after GPT-5.3 instant: Here’s what’s new","description":"OpenAI has launched its latest AI model, GPT-5.4, just days after rolling out GPT-5.3 Instant to ChatGPT users. The new model brings major improvements in reasoning, coding, and professional task automation across ChatGPT, the API, and Codex.","timestamp":1772797681,"category":"technology","headline":"OpenAI launches GPT-5.4 AI model days after GPT-5.3 instant: Here’s what’s new","modifiedHeadline":"OpenAI Unveils Advanced GPT-5.4 AI Model","subCategories":["scid_b7e786a9"],"subcategoriesName":["Artificial Intelligence"],"region":"IN","bundleMetaData":{"bundleId":"c35cfc57-1bba-4539-a24c-695ea99ac023","bundleTitle":"Mastering AI","cardData":null},"iabCategoryId":"597","iabCategoryName":"Artificial Intelligence","fcSummary":"openai has launched gpt-5.4, its latest ai model, focusing on professional tasks and enhanced reasoning. it integrates coding advancements and agent-based workflows for improved accuracy and speed, outperforming previous versions and even human experts in some areas. gpt-5.4 features better deep web search, reduced hallucinations, and faster coding capabilities, alongside a new tool search system to optimize efficiency.","summary":"OpenAI launches GPT-5.4, boosting reasoning, coding & automation. Model shows improved accuracy & outperforms experts in many tasks.","ctaUrl":"https://trends.glance.com/newz/astrology/IN/en/news9/f-73cd9b4b27-ca940bbf-efde-579d-a56b-d65e66013c74","sourceUrl":"https://www.news9live.com/technology/artificial-intelligence/openai-launches-gpt-5-4-ai-model-days-after-gpt-5-3-instant-heres-whats-new-2939755","bannerImage":"https://g-mob.glance-cdn.com/public/fc/image/WmUMkWTBK-Q0.webp","content":"

New Delhi: OpenAI has officially unveiled its latest artificial intelligence model, GPT-5.4, marking another rapid upgrade in the company’s GPT-5 series.

\n\n

The launch follows only days after the company introduced GPT-5.3 Instant to ChatGPT users, which is an indication of the rapid expansion of OpenAI towards enhancing AI-related functionalities to serve professional and consumer purposes alike.

The new model is launched on ChatGPT, API and Codex. The company states that GPT-5.4 is the first and most powerful and efficient frontier-based model to date, which integrates reasoning, coding, and agent-based workflow advancements to perform complex real-world tasks more accurately and faster.

\n\n\n

GPT-5.4 focuses on professional work and advanced reasoning

According to OpenAI, GPT-5.4 is the AI that should support users in doing complicated professional work better. The model incorporates the strength of the coding of GPT-5.3-Codex with better functionality in spreadsheets, presentations, documents, and other workplace applications.

\n
GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT.
\n
GPT-5.4 is also now available in the API and Codex.
\n
GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model. pic.twitter.com/1hy6xXLAmJ
\n
\n
— OpenAI (@OpenAI) March 5, 2026
\n

In ChatGPT, the model is available as GPT-5.4 Thinking, which can outline an upfront reasoning plan before generating its final answer. This enables the user to control the model in the middle of the response and optimise results more effectively without any numerous prompts. The model also has a better deep web search, and context is retained in the model during long or complex queries.

Improved accuracy and lower hallucinations

OpenAI says its most factual model is GPT-5.4 to date. According to internal analyses, the FD of individual assertions produced by the model is lower (33 per cent) than GPT-5.2. Factual errors are also 18 per cent less probable to be found in full responses.

\n\n

The model is also superior on the professional knowledge work tasks. In GDPval, which compares real-life tasks performed by GPT-5.4 with industry experts, the AI performed as well or better than human industry experts in 83 out of 44 occupations, compared to 70.9 with GPT-5.2.

The ability of GPT-5.4 to use computer environments is one of the largest upgrades, which enables AI agents to communicate with software environments. The model can use keyboard and mouse actions to browse, engage in the activities of programmes and accomplish tasks based on the screenshots.

\n\n

GPT-5.4 performed well in testing environments like OSWorld-Verified, with a high success rate of 75 per cent, outperforming even GPT-5.2 and, in certain cases, outperforming humans.

Stronger coding performance and faster developer workflows

OpenAI is also empowered by GPT-5.4 to code. The model both equips GPT-5.3-Codex on the SWE-Bench Pro benchmark and achieves lower latency in reasoning tasks.

Codex developers can also set a /fast mode which boosts token speed by up to 1.5 times. This enables the programmers to repeat the process of coding, debugging and testing working models in a shorter time but with the same level of model intelligence.

\n\n

Another new system introduced by OpenAI is the tool search, which enables GPT-5.4 to operate effectively in large systems of external tools. As opposed to having all tool definitions loaded into the prompt, the model is able to search and retrieve tools only on demand.

This will save a lot of time and minimise the use of tokens. When compared to testing in 250 trials, the tool search system minimised the use of tokens by 47 per cent without compromising on the accuracy.

In general, GPT-5.4 signifies a significant breakthrough on the journey of OpenAI towards more intelligent AI agents, expedited developer processes, and improved quality in ChatGPT, the API, and Codex.

\n","readDurationInMinutes":"6","author":{"name":"News9","logo":"https://glance-mob.glance-cdn.com/public/bubble/png/0e26af53-6a49-4ee1-a63d-25161b714ad5.jpeg"},"ism":false,"analyticsResponses":[{"measurementId":"G-EQBYL0ZYL8","type":"GA"},{"measurementId":"33425927","type":"COMSCORE","partnerName":"glance"}],"pointerSummary":["OpenAI launched GPT-5.4, a new AI model, days after GPT-5.3, expanding AI functionalities.","GPT-5.4 boasts improved factual accuracy (33% lower FD) and outperforms experts in 83/44 tasks.","Developers benefit from faster coding (1.5x speed) & token efficiency via new tool search system."],"tags":["OpenAI","GPT54","ArtificialIntelligence","AIModel","ChatGPT","API","Codex","AIAdvancements","Reasoning","Coding","ProfessionalTasks","AIAgents","FactualAccuracy","ReducedHallucinations","DeveloperWorkflows","GPT53Instant","GPT54Thinking","GPT54Pro","OSWorldVerified","SWEBenchPro"],"consent":true,"redirectionEnabled":false}

OpenAI launches GPT-5.4 AI model days after GPT-5.3 instant: Here’s what’s new

SUMMARY