On Tuesday, OpenAI released a new Responses API aimed at helping developers and businesses build autonomous AI agents using the company's models and frameworks. This toolkit effectively replaces OpenAI's Assistants API, which is slated for deprecation in the first half of 2026.
The Responses API allows businesses to develop custom AI agents capable of performing web searches, scanning company documents, and browsing websites, similar to OpenAI's Operator product. Developers can leverage the same AI models used in ChatGPT Search: gpt-40-search and gpt-40-mini-search, which can find answers to questions online and cite sources when generating responses.
In an interview, OpenAI product lead Olivier Gaudin stated: "Demo agents are very easy, but scaling agents is quite hard, and getting people to use them regularly is also quite hard." He hopes OpenAI can bridge the gap between AI agent demos and actual products this year, believing that "agents are the most impactful application of AI."
Alongside the Responses API, OpenAI also released an open-source toolkit called Agents SDK, providing developers with free tools to help integrate models with internal systems, implement safety measures, and monitor AI agent activity for debugging and optimization.
While OpenAI claims its search models boast high factual accuracy (gpt-40-search scored 90% in benchmark tests, and gpt-40-mini-search scored 88%), the company acknowledges limitations. Web search doesn't entirely solve the problem of AI hallucinations, and Computer-Using Agents (CUA) models "are not yet very reliable for automation tasks on operating systems."
