We’ve been expecting OpenAI to release something about agents for awhile now and we finally got that release today.
OpenAI enters the arena with a new Responses API and Agents SDK. They’re labeling this SDK as the production-level version of their previous framework Swarm.
OpenAI’s Agent Framework can work with other LLM providers using the OpenAI Chat Completions format (another plus for proxies like LiteLLM).
Three general categories of tools they call out:
- Hosted tools: these run on LLM servers alongside the AI models. OpenAI offers retrieval, web search and computer use as hosted tools.
- Function calling: these allow you to use any Python function as a tool.
- Agents as tools: this allows you to use an agent as a tool, allowing Agents to call other agents without handing off to them.
Kind of cool how they are turning Python functions into Agent tools:
- The Agents SDK will setup the tool automatically:
- The name of the tool will be the name of the Python function (or you can provide a name)
- Tool description will be taken from the docstring of the function (or you can provide a description)
- The schema for the function inputs is automatically created from the function’s arguments
- Descriptions for each input are taken from the docstring of the function, unless disabled
- We use Python’s inspect module to extract the function signature, along with griffe to parse docstrings and pydantic for schema creation.
Other notes I took from watching the above release video:
OpenAI has a fine-tuned version of GPT-4o that they use to review search results that are coming back from the Search tool in ChatGPT. It out performs o3-mini and gpt-4.5 on SimpleQA. This model is used in the Search tool they are making available.
Other tools are a File Search tool and Computer Use tool (which it sounded like also has a different fine-tuned CUA model?). Actually yes, there is a computer use model. One of them is computer-use-preview-2025-02-04 (see here).
They’re also releasing a Responses API. OpenAI released the Chat Completions API in March 2023 with gpt-3.5-turbo. It will be interesting if this Responses API becomes the de-facto standard like Chat Completions has. They will continue to support Chat Completions but Responses API will be a superset of Completions API features.
The Assistants API will eventually be deprecated sometime in 2026.