Batch vs Flex vs Standard vs Priority — same model, four prices.
Latency is the lever: async Batch work should never pay standard price, and only truly interactive paths need Priority. Sizing a single model? Rank every model on the 300+ model comparison, and check async savings on the batch API savings calculator.
Free LLM service tier calculator — your tokens and standard price to the cost on Batch, Flex, Standard and Priority tiers.
Most big providers sell the same model at several latency tiers. Batch runs your requests asynchronously (often within 24h) for about half price; a Flex or service tier is cheaper but slower or variable; Standard is the default; a Priority tier costs a premium for guaranteed fast responses.
Whenever the work is not user-facing in real time — bulk summarisation, embeddings, evals, data labelling, overnight jobs. Batch is typically ~50% off, so moving async workloads off the standard tier can roughly halve that part of the bill for zero quality loss.