Methods: client.create_batch(), client.get_batch(), client.cancel_batch(), client.get_async_job_websocket_url("batch", batch_id, ...), client.batches.websocket_url(batch_id, ...).
batch = client.create_batch({
"endpoint": "responses",
"input_file_id": "file_123",
"completion_window": "24h",
"session_id": "agent-run-42",
})
status = client.get_batch("batch_123")
cancelling = client.cancel_batch("batch_123")
websocket_url = client.batches.websocket_url("batch_123", interval_ms=1500)
print(batch)
print(status)
Key parameters
endpoint (required): Target endpoint for batch items such as responses.input_file_id (required): File id uploaded through the files API.completion_window: Upstream batch completion window such as 24h.session_id: Optional AI Stats grouping id for logs, sessions, and investigate tooling.webhook: Optional webhook configuration for async lifecycle notifications.metadata: Optional object persisted with the batch.
Batch responses also surface gateway observability fields such as request_id, provider, echoed session_id / webhook, and terminal billing / pricing_lines when available.
Use client.batches.websocket_url(...) when you want to subscribe to the documented /v1/async/batch/{id}/ws lifecycle stream instead of polling only.
Use client.get_async_job_websocket_url("batch", batch_id, ...) when you already have a generic async job kind/id pair and do not want to go through the resource helper. Last modified on May 6, 2026
