According to a recent LinkedIn post from OpenRouter, the company is highlighting the availability of a Batch API aimed at workloads that do not require immediate responses. The post suggests this batch capability can cut per‑token pricing roughly in half on most models while returning results within a 24‑hour window, often much faster.
The LinkedIn content indicates the Batch API supports lists of requests via a single POST and returns outputs in the same JSON format used for chat completions, responses, Anthropic messages, or embeddings. OpenRouter’s post cites beta data from more than 230,000 batches, with a median completion time of about seven minutes and 90% reportedly finishing within an hour.
The post also notes that submission timing affects tail latency, with batches sent outside 5 a.m. to noon Pacific reportedly seeing even the slowest 10% complete in about an hour. Performance appears to scale favorably, with one request finishing in roughly 5–11 minutes and 1,000 or more requests taking about 12–21 minutes, depending on submission time.
Feature details in the post include per‑batch provider selection based on the cheapest eligible endpoint, optional provider pinning, and bring‑your‑own‑key routing with separate BYOK fees. Results are keyed by user‑defined IDs, designed so partial failures do not invalidate an entire job, and are returned inline via simple JSON over two calls, with all batches logged by model, provider, status, and cost.
The post positions this Batch API for use cases such as corpus labeling, backfilling embeddings, evaluation scoring, ticket‑backlog summarization, and large‑scale prompt runs across thousands of rows. For investors, this focus on lower‑cost, high‑throughput batch processing across more than 70 models could enhance OpenRouter’s appeal to price‑sensitive enterprise users and data teams.
If adoption scales, the offering may support higher volume utilization of OpenRouter’s infrastructure while reinforcing its role as an aggregation layer over multiple AI providers. However, the LinkedIn post does not disclose pricing details beyond relative discounts or indicate how monetization and margins will evolve, so the financial impact remains uncertain and will depend on actual customer uptake and usage intensity.