OpenAI has unveiled a new Responses API, aimed at simplifying the creation and deployment of AI agents capable of independently performing tasks. This latest offering is set to replace the existing Assistants API, which will be phased out within a year.
The Responses API enables developers to build AI-powered agents that can search company datasets and browse the internet, significantly enhancing accuracy and real-time information retrieval. This aligns with OpenAI’s broader vision of integrating AI agents into everyday workflows, making automation more seamless.
Enhanced Search Capabilities and AI Models
When developing AI agents with the Responses API, developers can choose between two models: GPT-4o Search and GPT-4o Mini Search. Both models are designed to autonomously browse the web and cite sources, addressing one of the biggest challenges in AI—hallucinations and factual errors.
OpenAI demonstrated the effectiveness of these search-capable models using its SimpleQA benchmark, which evaluates AI confabulation rates. The results:
- GPT-4o Search scored 90%, making significantly fewer errors than previous models.
- GPT-4o Mini Search followed closely at 88%.
- GPT-4.5, despite having more parameters, only scored 63% due to its lack of search capabilities.
While these improvements are substantial, OpenAI acknowledges that the models are not infallible—GPT-4o Search still produces factual mistakes in around 10% of responses. This level of error may be too high for some critical AI applications, but the company expects reliability to improve over time.
New Tools for Developers
To further drive AI adoption, OpenAI has released an open-source Agents SDK, equipping developers with tools to integrate AI models into internal systems while implementing safeguards and monitoring mechanisms. Additionally, OpenAI has introduced Swarm, a framework for managing and orchestrating multiple AI agents simultaneously.
This strategy mirrors OpenAI’s previous move of embedding ChatGPT into Apple’s Siri through Apple Intelligence, exposing its models to a broader user base. Industry experts believe the Responses API will further accelerate public adoption of AI agents across various software platforms.
AI Agents Still a Work in Progress
Despite OpenAI’s advancements, AI agents remain in their early stages. While some developers are eager to explore new capabilities, recent events—such as the highly anticipated Chinese AI agent Manus, which initially impressed early adopters but later fell short—serve as a reminder that these technologies are still evolving.
Stay tuned to DC Brief for further updates on this story and other technology developments.
