OpenAI Agents SDK 0.25.0 released
A minor release that removes the Runloop sandbox provider, improves checkpoint handling for nested approvals, and updates model defaults.
What changed in 0.25.0
- Runloop sandbox provider removed; users must migrate to another supported provider like Modal before upgrading.
- Pydantic run contexts now preserve default values and in-place edits in checkpoints; custom serializers may be needed for process-local fields.
- Session compaction now defaults to gpt-6-luna model instead of gpt-4.1; set model="gpt-4.1" to retain previous behavior.
- MCP dependency raised to version 1.30.0 or later; nested tool approval decisions now preserved across resume with output guardrails.
- Multiple bug fixes for tool context approvals, concurrent HTTP requests, completed tool call fingerprinting, and handoff history handling.
Check before you update
- Runloop sandbox provider and runloop extra removed; switch to Modal or other supported sandbox providers.
- OpenAIResponsesCompactionSession default model changed from gpt-4.1 to gpt-6-luna.
- MCP dependency minimum raised from 1.21.2 to 1.30.0.
Summarised by AgentGid from OpenAI's official release notes on GitHub, as reported there; read the original before relying on it. Price, status and release dates come from our daily data.
OpenAI Agents SDK at a glance
| What it is | OpenAI's lightweight framework for multi-agent workflows, with handoffs, guardrails, sessions and tracing. Available for Python and TypeScript. Per its PyPI page it is provider-agnostic, supporting the OpenAI Responses … |
|---|---|
| Latest version | 0.25.0 is the latest stable release we track. |
| Release pace | 1 stable release in the last 30 days; 0.25.0 came 34 days after the previous one |
| Price | Free (OSS). OpenAI Agents SDK is open source and free to use; you pay only for the models and infrastructure it runs on. (checked Nov 4, 2026) |
| Downloads, last 7 days | 7.1M up 3.5% |
| Gid Score | 61 #6 in Agent frameworks and SDKs |