On Tuesday, OpenNai released a new tool designed to help developers and companies build Agent AI – Automatic systems that can complete the company model and framework.
The device is part of the new response Openenai API, which can use a special AI agent that can take a web search agent, scan through corporate files, and navigate websites, and navigate websites Opeain carrier productsSee rankings-. API response effectively replace the opening API ASSISTThe company planned to be the sun in the first half of 2026.
Hype around AI Agent has grown dramatically in the past-year-year-year-year industry fighting for others, or even determinesWhat is “Ai Ai Ai”. In the latest examples of Hype Agents first in front of the utility, the early chinese effects of the week are viruses For a new Agent platform called the Manus The user is hurriedly found not to deliver the company.
With other words, the water is high for the opening to earn an agent.
“Very good to post your agent,” Olivier Godemont, Opiai Product Head, tells the techcrunch in the interview. “For a scale of agents quite difficult, and let people use it often.”
Before this year, Opening introduced a Agent AI in Chatgpt: Operator, navigating websites for you, and Inside deep researchThe compilation of research reports for you. The two devices give sexy in what the agent of agents can do, but it is still left easily in the “Autonomous” department.
Now with the API response, Opening wants to sell access to the components that conduct A AI Agent, allow the developer to build your own carrier- and application of deep research agent. Openai hope that developers can make some applications with a preferred agent technology for now.
Using Fire Responses, developers can tap the same AI model (in preview) above the opening Search Chatgpt Search Web Web: GPT-4O Search and mini 4o Gpt-4O Search. The model can browse the web to the answer to the question, quote the source when you produce reply.
Openai claimed that GPT-4O Search and a very precise mini-4O mini search. The company’s simple benchmark, that measures a brief model ability, Fact-arrested 90% Score selection when GPT-4O search score 88% (better better). To compare, Gpt-4.5 – Opening a larger model, recently released – score only 63%.
The fact that the AI fighting tools are more accurate than the traditional AI AI model should not be surprised – search GPT-4O can only find the right answers. However, the web search does not translate halutos that has been resolved. Beyond the facts of the fact, AI search tool is also likely to struggle short, such as “Lakers score today”), and new report shows ChatPT Quotation is not always reliableSee rankings-.
API response also includes searching services proportion files that can scan files in corporate databases to retrieve information. (The opening states that it does not proclaim the model in the file. The model produces mouse and keyboard acts, the developer operators using the computer like a data store and app app.
Companies can choose CUA models, they release the research previews, local system, the opening system. Version CUA consumer available on the operator can only act on the web.
So clearly, the API response will not solve all the technical issues that plaguing the AI AI today.
While AI selections are used more accurate than traditional ai ai Ai models – the fact that they are granted that they cannot find the correct answer – Web search does not give AI Halulnation of the issues that have been completedSee rankings-. GPT-4O is still getting 10% of the wrong question. Beyond accuracy, AI search tool is also likely to be Struggle with short navigation questions (eg “Lakers score today”), and new report shows ChatPT Quotation is not always reliableSee rankings-.
In blog postcrunch provided to TechCrunch, Openai said that CUa model “did not trust the task of the duty in the operating system,” and it’s easy to make mistakes “.
However, the opening says this is an excessive agent tool, and continues to be used to improve.
Together API response, the opening launches the Open-source-source offers, who offers developers to integrate internal models, and monitoring AI agents for debugging. SDK Agency is a variety of multi-opery gang, framework for companies released at the end of the last year.
Godemont said that the opening was able to chase the gap between the tour and the AGA product of this year, and if the agent is the largest AI application that will happen. ” That convey the opening proclamation of the CEO Sam Altman was created in January: 2025 is an AI Acy Agent entered the labor force.
Whether 2025 really became “A AI AI year,” The latest Openai Release shows the company you want the shift from the Flashy Agent tour for an affective tool.
