Developers who use API responses can access the same models in which Power Chatgpt: GPT-4O Search and GPT-4O Mini Search. These models can Browse the web To answer questions and cite sources in their answers.
This is noticeable because Openai says that the ability to search the added web works greatly to improve the realistic accuracy of artificial intelligence models. On Openai Simpleqa StandardWhich aims to measure the defect, the GPT-4O 90 percent search recorded, while the GPT-4O MINI search has achieved 88 percent-he greatly outperforms the GPT-4.5 larger model without research, which recorded 63 percent.
Despite these improvements, technology still has great restrictions. Regardless of CUA problems properly navigating on web sites, the improved search is not completely solved by the Confabulations AI problem, with the continued search in GPT-4O on real errors of 10 percent of the time.
Along with API responses, Openai released an open source SDK agentsProviding developers with free tools to integrate models with internal systems, implement protection, and monitoring agent activities. The tools group follow the previous Openai version of swarmA framework for organizing many agents.
These first days are still in the field of artificial intelligence agent, and things are likely to improve quickly. However, at the present time, the movement of artificial intelligence agent remains vulnerable to unrealistic demands, as shown earlier this week when Discover users The effect of the emerging Chinese butterfly Manus you have The agent platform has failed to achieve many of its promises, while highlighting the ongoing gap between promotional claims and practical functions in this emerging category in the field of technology.