Streaming tool calls
Tool calls now stream partial arguments and results, cutting perceived latency on multi-step runs by about 40%.
What changed
Tool calls stream partial arguments while the model is still writing them, and results stream back as each tool returns.
Run traces now show time to first token for every tool step. Perceived latency on multi-step runs drops by about 40%.
None. Set stream: true on any run to opt in. Existing clients keep working without changes.
Release notes
Agents now stream tool calls as they happen. Partial arguments arrive while the model is still writing them, and results stream back as the tool returns them.
Set stream: true on any run. Existing clients keep working without changes.
Questions about this release
The docs cover every flag and endpoint. For anything else, support answers within one business day.
Run your first agent in five minutes.
1M free tokens a month. No card, no sales call, production keys in two minutes.