OpenAI has announced the launch of new tools designed to facilitate the creation of AI agents. The release includes a Responses API, various integrated tools (web search, file exploration, and computer action execution), and Agents SDK, an open-source tool for orchestrating workflows with artificial intelligence.
With these “building blocks”, the company led by Sam Altman aims to simplify the process of building, deploying, and scaling autonomous agents, responding to the demand from developers and businesses for more efficient solutions to incorporate AI into their operations. This milestone promises to revolutionize the capability of agents to search for information, manage data, and automate complex tasks, marking a new phase in the development of advanced AI applications.
The Responses API by OpenAI is a new interface designed to facilitate the creation of AI agents through integrated tools. It combines the simplicity of the Chat Completions API with the ability to use advanced tools from the Assistants API, enabling developers to manage complex tasks in a single call. Additionally, it includes usability improvements, such as a unified element-based design and optimized access to the model’s text output.
This API supports tools such as web search, file search, and computer usage, connecting models with the real world to perform more advanced tasks. It also facilitates the evaluation of agent performance without the need to integrate multiple APIs or external providers. There is no additional cost, as it is billed per the usage of tokens and tools according to standard rates.
Regarding existing APIs, Chat Completions will remain compatible and will be updated with new models for those who do not require integrated tools. However, the Responses API is a more advanced and flexible version, recommended for new integrations. OpenAI plans to discontinue the Assistants API by mid-2026, integrating its functionalities into the Responses API and providing a migration guide to facilitate developers’ transition.
Next, we explain in detail each of the integrated tools of the Responses API:
Agents can access up-to-date information in real-time with web search, which is available in GPT-4o and GPT-4o-mini. This tool has been applied in shopping assistants, research, and travel reservations, providing answers based on verified sources. Companies like Hebbia use it to extract information from large data volumes, improving financial and legal analyses. Its accuracy has been validated in SimpleQA, with scores of 90% in GPT-4o and 88% in GPT-4o-mini.
The tool is in a preliminary version and allows developers to integrate searches directly into their applications, with prices starting from $33 per thousand queries.
It allows the retrieval of information from large volumes of documents, with support for various formats, query optimization, and metadata filtering. Applications like Navan use it to query business policies in travel agents, reducing search times and customizing responses for each user.
The tool costs $2.75 per thousand queries, with storage at $0.11/GB/day (the first GB is free).
This function, based on the CUA (Computer Use Agent) model, allows agents to execute tasks in digital environments, from workflow automation to data entry in systems without an API. Companies such as Unify and Luminai have utilized this technology to enhance operational efficiency in sales and legacy system integration.
The tool is in a research phase. It is available for advanced developers with a cost of $3 per million input tokens and $12 per million output tokens. To minimize risks, OpenAI has implemented security measures such as checks against code injections and human supervision in critical tasks.
These tools expand the potential of AI agents, connecting them with real-time data and enabling them to act more effectively in digital environments.
Along with the Responses API, OpenAI has introduced the Agents SDK. This is a tool designed to facilitate the orchestration of workflows involving one or more agents. This agent allows developers to efficiently coordinate interactions between different agents, ensuring they work together to achieve common goals. This capability is especially useful in scenarios where multiple agents must collaborate to solve complex tasks that require various skills and resources.
Compared to Swarm, its previous experimental version, it incorporates key improvements in agent management, including:
This SDK is useful for applications such as automation of customer service, content generation, code review, and sales prospecting. Companies like Coinbase have used it to develop AgentKit, a system that allows AI agents to manage cryptocurrency wallets in just hours. Meanwhile, Box has integrated it to search for information in internal data and public sources, facilitating decision-making in the financial sectors.
It is compatible with the Responses API and the Chat Completions API. This SDK can be integrated with models from other providers. It is also compatible with Python and soon with Node.js. OpenAI will continue its development as an open-source project, promoting community collaboration to expand its functionalities.
Photo: OpenAI
Your email address will not be published. Required fields are marked *
Δ