Source: OpenAI Batch API Guide
OpenAI API guide to the Batch API: JSONL input files, 24-hour window, 50% discount, limits, statuses and expiry behaviour.
Contents
- What: official API documentation page
- Author/publisher: OpenAI
- Date: living document; read 2026-11-14
- Original: Batch API
Summary
OpenAI batches are created from an uploaded JSONL file in which each line is a full request. Jobs complete within a 24-hour window at 50% of synchronous price and draw on a separate rate-limit pool.
Key claims
- Input line:
custom_id,method,url,body; one model per file; uniquecustom_ids. - Endpoints:
/v1/responses,/v1/chat/completions,/v1/embeddings,/v1/completions,/v1/moderations, image generation and edits. - Limits: 50,000 requests and 200 MB per input file; 2,000 batch creations per hour; per-model queued-token limits.
- Statuses:
validating,failed,in_progress,finalizing,completed,expired,cancelling,cancelled. - Results go to an output file, failures to a separate error file; order is not guaranteed.
- On expiry, unfinished requests fail with
batch_expired; completed results remain and are billed. - Output files expire after 30 days by default (configurable 1 h–30 d). Cancelling can take up to 10 minutes.