Batch API: half-price inference by bundling requests

OpenRouter's new Batch API lets you submit a workload together for results within 24 hours, typically at half the normal per-token price. It supports more than 70 models.

OpenRouter reports a median completion time of 7 minutes across 230,000+ batches in its two-week beta. That's a reported result, while the processing window is 24 hours.

Batches support chat completions, responses, messages, and embeddings. Results return independently, so bad rows don't fail the whole job. Images and files must use public URLs. Audio, video, and OpenRouter's web search plugin aren't supported; web search calls that do run bill at standard rates. Inputs and results are retained for 30 days or until deleted.