Source: Gemini Batch API Docs

Google Gemini API documentation on the Batch API: 50% cost, 24-hour target, inline vs JSONL file input, job states.

Contents
  • What: official API documentation page
  • Author/publisher: Google
  • Date: living document; read 2026-11-14
  • Original: Batch API

Summary

The Gemini Batch API (generateContent only) processes large volumes of requests asynchronously at 50% of standard price, with a 24-hour target turnaround. Requests are submitted inline or as a JSONL file.

Key claims

  • 50% of standard cost; target turnaround 24 hours, usually much faster; meant for non-urgent work such as preprocessing and evaluations.
  • Inline requests suit small batches under 20 MB total; larger jobs use a JSONL file (up to 2 GB, uploaded via the File API) of GenerateContentRequest objects, with JSONL output of responses or status objects.
  • Job states: JOB_STATE_PENDING, running, JOB_STATE_SUCCEEDED, JOB_STATE_FAILED, JOB_STATE_CANCELLED, JOB_STATE_EXPIRED (pending or running for more than 48 hours, no results).
  • The docs present higher rate limits as part of the trade for up-to-24-hour turnaround, and batch jobs also support the embeddings model for higher throughput.
  • Batch pricing applies to cached tokens too (e.g. Gemini 3.8 Flash batch cached input $0.0375/MTok vs $0.075 standard, through 2026), with the same hourly storage charge. The batch guide itself says explicit caches are referenced per request via cached_content and that cache hits are billed “at standard context caching rates”, which reads differently from the pricing page’s batch-tier cached rate.

Feeds

Batch processing

Suggest changes →