What's the difference between the OpenAI Agents SDK and Google's Agent Development Kit (ADK)?

The OpenAI Agents SDK runs OpenAI’s Responses API by default and reaches other model providers only through its LiteLLM or Any-LLM adapter packages, both still in beta; Google’s Agent Development Kit (ADK) is model-agnostic from the start, reaching Claude through a native, direct integration and GPT natively only in some ADK languages (Go ships its own OpenAI package), through the LiteLLM connector elsewhere, including Python. The two also organize work differently. The OpenAI SDK stays close to one agent loop, with a handoff passing the conversation to another agent and a guardrail checking its input or output. ADK is built around wiring several agents into named workflow types: Sequential, Parallel, and Loop for Java, graph-based and dynamic workflows since ADK 2.0 for Python, TypeScript, and Go.

That split matches the Claude Agent SDK’s own line against ADK: a single-model-first SDK built around one agent’s loop, against a model-agnostic framework built around a workflow shape. Picking the OpenAI Agents SDK means picking OpenAI’s models and a thin loop first; picking ADK means picking a workflow structure and a model second.