Batch Api

Agent submits 5000 requests to POST /v1/messages/batches, each carrying a unique custom_id matching its row in the operator's dataset.

Build requests[] with custom_id and params (model, messages, max_tokens, tools, etc.) per row. custom_id is the only way to map results back to source rows — pick a stable, unique value (e.g., row_uuid).

Batch status transitions: processing → ended (24h max latency window). Agent polls every 10 minutes.

Poll GET /v1/messages/batches/{id} with exponential backoff (start ~30s, cap at minutes); rely on processing_status='ended' as the terminal state. Do not assume completion before the 24h SLA — partial-complete batches remain 'processing' until either fully done or expired.

Batch status=ended. Agent fetches GET /v1/messages/batches/{id}/results and expects a JSONL stream.

Stream JSONL line-by-line; each row has custom_id + result (succeeded with message body | errored with error info | canceled | expired). Results URL remains valid for 29 days — persist results to your own store within that window if needed for replay.