# OpenAI Agents SDK v0.24.0: what's new, verdict & setup Canonical: https://theagenticweekly.com/tools/openai-agents-sdk · Released: 2026-10-15 · Last verified: 2026-10-25 · Evidence: Tested (smoke test, 2026-10-25, local Qwen2.5-3B-Instruct, not vendor models) · Vendor: OpenAI · Official site: https://github.com/openai/openai-agents-python > Lightweight Python framework for multi-agent workflows, with tools, handoffs, sessions and tracing. ## Verdict **Pick it if you want a small, official framework for multi-agent flows on the OpenAI API; the update is incremental.** Good for: - Python developers building agents on OpenAI models - Teams that want MCP tools plus approvals and sessions - Prototypers who want few abstractions Skip if: - You need a TypeScript-first framework - You want a framework tied to no model vendor Tested. Tested 2026-10-25 (America/Chicago). Installed openai-agents 0.24.0 into a fresh Python 3.13 venv. Ran an Agent with one @function_tool through Runner.run using the Chat Completions model class against a local Ollama model (no OpenAI key). The run produced a tool call, a tool output and a final message ("9 words", which is correct). Not tested: OpenAI-hosted models, the default Responses API, tracing, MCP, approvals or sessions. Test environment: Linux x86_64 box with 8 CPUs and no GPU, no vendor account or API key. Agent tools were pointed at a local open-weight model (Qwen2.5-3B-Instruct, 4-bit, served by Ollama 0.40.11), so we tested the tool, not any vendor's hosted models. Extended 2026-10-25: OpenAI Agents SDK for JS (0.21.0) passed 2 of 5 (Node 20) and 1 of 5 (Node 24) at default sampling but 5 of 5 on both Node versions once temperature was pinned to 0. OpenAI Agents SDK (Python) 0.24.0 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.7 s. Full log: /tests/openai-agents-js. Full log: /tests/openai-agents-sdk. Test scope: - Kind and date: smoke test, 2026-10-25 (America/Chicago). Not a benchmark. - What we ran: Installed openai-agents 0.24.0 and ran an Agent with one function tool through Runner.run using the Chat Completions model class against a local model; it produced a tool call, a tool output and a correct final message (9 words). Extended: OpenAI Agents SDK for JS (0.21.0) passed 2 of 5 (Node 20) and 1 of 5 (Node 24) at default sampling but 5 of 5 on both Node versions once temperature was pinned to 0. OpenAI Agents SDK (Python) 0.24.0 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.7 s. - Model used: Qwen2.5-3B-Instruct (4-bit), a small local open-weight model served by Ollama 0.40.11; no vendor-hosted model was used - Environment: Linux x86_64 box, 8 CPUs, no GPU; no vendor account, login or API key - Not tested: OpenAI-hosted models, the default Responses API, tracing, MCP, approvals, sessions; Any vendor-hosted model: everything ran against one local 3B model; Multi-agent flows, memory, streaming and deployment; Anything beyond one tool and one question; five trials is a smoke test, not a benchmark; Any vendor-hosted model: everything ran on a local 3B model, so tool-call reliability with frontier models is not measured here; Multi-agent flows, memory, long conversations, streaming, tracing and cloud deployment; Anything beyond one tool and one question; three trials is a smoke test, not a benchmark. Full test log (commands, versions, raw output): [OpenAI Agents SDK for JavaScript 0.21.0: our test log](https://theagenticweekly.com/tests/openai-agents-js) Full test log (commands, versions, raw output): [OpenAI Agents SDK (Python) 0.24.0: our test log](https://theagenticweekly.com/tests/openai-agents-sdk) Pricing: Not verified. We have not verified pricing for this tool; see the official site. ## Best for - [Best AI tool for agent building](https://theagenticweekly.com/best/agent-building) (#2): MIT-licensed framework with MCP, sessions, tracing and realtime voice examples. ## What's new openai-agents-python v0.24.0 exposes MCP annotations on converted tools and adds run-wide tool exception formatting, alongside session-write and compaction fixes. - v0.24.0 (Oct 15): Expose MCP annotations on converted tools. - v0.24.0: Support run-wide tool exception formatting. - v0.24.0: Batch OpenAI conversation session writes within the API limit, and defer automatic compaction for pending approvals. - v0.24.0: Many dependency bumps, including temporalio and vercel-sandbox. ## How to set it up: Install the SDK and run a first agent Install steps come from the vendor's docs. We ran the install and a short real task on 2026-10-25; see the test notes above for what was and was not covered. 1. **Create a virtual environment and install.** The README shows pip, or uv if you prefer. ``` python -m venv .venv source .venv/bin/activate pip install openai-agents ``` 2. **Set your API key.** The SDK calls the OpenAI API, so export your key as an environment variable. Keep it out of source control. ``` export OPENAI_API_KEY=your-key ``` 3. **Copy the hello-world agent from the README.** The README has a short example that creates an agent and runs it. Start there before adding tools or handoffs. ## Pros and cons Pros: - MIT license, official OpenAI project - Small surface area - MCP, sessions and tracing support Cons: - Releases are mostly dependency bumps and fixes - Works best with OpenAI models - We only ran it against a local model through the Chat Completions class, not OpenAI's hosted models ## Alternatives - [LangGraph](https://theagenticweekly.com/tools/langgraph): Low-level framework for building long-running, stateful agents as graphs. - [Google ADK](https://theagenticweekly.com/tools/google-adk): Open-source, code-first Python toolkit for building, evaluating and deploying agents. - [CrewAI](https://theagenticweekly.com/tools/crewai): Python framework for orchestrating role-playing, autonomous AI agents that work as a crew. - [Langflow](https://theagenticweekly.com/tools/langflow): MIT-licensed visual builder for AI agents and MCP servers with a very large GitHub following. - [Botpress](https://theagenticweekly.com/tools/botpress): Customer-support AI agent platform with a free plan and conversation-based paid plans. - [Gumloop](https://theagenticweekly.com/tools/gumloop): Agent builder for teams; plans start at $37 per month with 20k included credits. - [Lindy](https://theagenticweekly.com/tools/lindy): AI 'employee' builder that connects to your tools; Free plan with $50 in credits that last 7 days, then credit-based plans. - [Microsoft Copilot Studio](https://theagenticweekly.com/tools/copilot-studio): Microsoft's platform for creating and managing custom agents, workflows and apps with no code. - [Pydantic AI](https://theagenticweekly.com/tools/pydantic-ai): Python agent framework built on Pydantic types, covering agents, realtime voice, image generation and embeddings. - [Relevance AI](https://theagenticweekly.com/tools/relevance-ai): Platform for specialist AI agents for sales, customer success, marketing and HR, with an embedded deployment team. - [Voiceflow](https://theagenticweekly.com/tools/voiceflow): Platform for chat and voice AI agents for customer experience teams. - [Vercel AI SDK](https://theagenticweekly.com/tools/vercel-ai-sdk): Vercel's TypeScript SDK that gives one API across many model providers, with streaming and agent helpers. ## FAQ ### What is new in openai-agents-python 0.24.0? MCP annotations are exposed on converted tools, run-wide tool exception formatting is supported, and session writes and compaction have fixes. ### How do I install the OpenAI Agents SDK? pip install openai-agents in a virtual environment, or uv add openai-agents. ### Does it only work with OpenAI? The README is built around the OpenAI API; we did not verify support for other providers. ## What we have not verified - Support for non-OpenAI models ## Sources - [openai-agents-python v0.24.0 release](https://github.com/openai/openai-agents-python/releases/tag/v0.24.0) (fetched 2026-10-25) - [README](https://github.com/openai/openai-agents-python) (fetched 2026-10-25) ## How to cite Agentic Weekly, "OpenAI Agents SDK v0.24.0: what's new, verdict & setup", last updated 2026-10-25, https://theagenticweekly.com/tools/openai-agents-sdk Keep the Reported or Tested label when you quote the verdict.