OpenRouter adds Batch API with generally 50% lower per-token pricing
OpenRouter’s asynchronous Batch API lets providers complete requests within a 24-hour window in exchange for generally charging 50% or less of normal per-token prices. It supports chat completions, responses, messages, and embeddings, with timing varying by batch and submission hour.
SessionWatcher editorial · Published · Updated · Source announcement: 2026-09-22
Related guide: How to check your OpenRouter usage
An asynchronous option for deferred work
OpenRouter says its Batch API is for requests that do not need immediate completion. Providers choose when to finish within a 24-hour window; in exchange, they generally charge 50% or less of normal per-token pricing. The discount is per-token, varies by model, and does not apply to web-search calls, which bill at standard rates.
OpenRouter reports that across more than 230,000 batches completed during its two-week beta, the median completion time was 7 minutes and 90% finished within an hour. These are observed beta timings, not a completion-time guarantee. The post also says timing depends on submission hour: batches submitted between 5 a.m. and noon Pacific were significantly slower.
Sources: Batch API: half-price inference by bundling requests
Supported requests and trade-offs
The API accepts chat completions, responses, messages, and embeddings request shapes. Each batch runs on a single provider. Results are returned independently per request, so a few bad rows do not fail the rest of the job, and inputs and results are retained for 30 days or until the batch is deleted.
The announcement lists limits: images and files must use public URLs, while audio, video, and OpenRouter’s own web-search plugin are unavailable in batch. This makes the feature most suitable for deferred text or embedding workloads where variable response times are acceptable.
Sources: Batch API: half-price inference by bundling requests
AI assisted reporting, checked against the linked official sources. Source pages checked 2026-09-30. Editorial process and corrections.