Batch API
An asynchronous processing mode that trades immediacy for a large discount — commonly 50% off input and output.
A Batch API lets you submit many requests at once and collect the results asynchronously, usually within a stated window such as 24 hours, in exchange for a substantial discount. The standard discount is 50% off both input and output, offered by Anthropic, OpenAI, Google, and Mistral among the providers in this explorer.
Batch mode is ideal for work that is not latency-sensitive: bulk classification, embedding-style enrichment, offline summarization, evaluation runs, and dataset generation. Because the provider can schedule the work when capacity is available, it passes the savings back to you. The trade-off is that you cannot use batch for anything interactive — there is no real-time response.
On this explorer, where a provider offers batch but does not publish a separate batch column, the batch rate is derived as exactly 50% of the standard rate — the documented convention — and labeled as such. Providers without a batch offering show a dash, never a derived number. If a large share of your volume can tolerate a delay, moving it to batch is often the simplest way to halve that portion of your bill; model it with the batch toggle in the hub calculator.
FAQ
How much cheaper is the Batch API?
Typically 50% off both input and output tokens, for requests you can process asynchronously rather than in real time.
What is the catch with batch pricing?
Latency. Batch requests are processed within a window (often up to 24 hours), so the mode suits offline, non-interactive workloads — not live chat or agents.