OpenAI has announced the new Responses API, the Agents SDK, and observability tools to address the challenges that creating production-ready agents pose, such as building custom orchestration, and handling prompt iteration across complex, multi-step tasks.
OpenAI says agents will soon become integral to the workforce, enhancing productivity across industries, by accomplishing complex tasks using advanced capabilities such as reasoning and multi-modal interactions. The new tools launched by OpenAI aim to make it easier for developers to build agentic workflows using the OpenAI platform.
The Responses API combines chat completions with assistant capabilities and should take precedence over the Chat Completions API and the Assistants API for new projects.
We believe the Responses API will provide a more flexible foundation for developers building agentic applications. With a single Responses API call, developers will be able to solve increasingly complex tasks using multiple tools and model turns.
The new API also provides built-in support to external tools including web search, local file search, and computer control using mouse and keyboard. Additionally, it provides several improvements that will make its use more straightforward based on developer feedback on the previous models, including a unified design, simpler polymorphism, improved streaming, and several SDK helpers.
For Web search, the Responses API leverages the same models used for ChatGPT search, GPT-4o search preview and GPT-4o mini search preview. The models scored 90% and 88% accuracy on the SimpleQA benchmark, which is significantly better than “plain-vanilla” GPT models performance (comprised between 15% and 63%). However, the computer use tool fares at 38.1% on the OSWorld benchmark, which suggests the model is not yet highly reliable for automating tasks on operating systems.
While both the Chat Completions API and the Assistants API will remain available for now, with OpenAI committed to enhancing the former with new models and features, the company has already announced that the Assistants API will be deprecated next year.
Along with the Responses API, OpenAI has also launched the new Agents SDK aimed at orchestrating agentic workflows by defining distinct agents, managing control transfer between them (handoffs), defining safety checks for input and output to prevent irrelevant, harmful, or undesirable behavior (guardrails), and visualizing traces to observe agents.
The Agents SDK is suitable for various real-world applications, including customer support automation, multi-step research, content generation, code review, and sales prospecting.
The Agents SDK supports all current OpenAI models, including o1, o3-mini, GPT-4.5, GPT-4o, and GPT-4o-mini. It also allows developers to augment their agents with external and persistent knowledge represented through vector stores and the Embeddings API. Relying on the Responses API, the Agents SDK supports the same external tools to search the Web, local files, or control a computer.
The Agents SDK supersedes its experimental orchestration agent Swarm and is compatible with any Chat Completions-style API, including the Responses API and third-party APIs.
Among community reactions, some Hacker News (HN) readers expressed their sentiment that OpenAI’s move away from the Chat Completions API is dictated by “non-technical” reasons and brings the risk of lock-in with their platform. In the same line, some readers suggest that the phasing out of the Assistant API indicates that a good approach is not rewriting code for the Responses API, but rather creating a wrapper so you have a chance to replace the underlying LLM in case you need it.
On another front, several HN readers pointed out that adopting the Agents SDK or any other agentic middleware means you are basically “outsourcing” your state management and business logic to a third party whereas you may prefer to keep the LLM-component as small as possible and build your own logic around it.
