Batch API Savings: Half Price for Patient Workloads
If a job does not need an instant answer, the batch API is close to free money: providers commonly halve token prices in exchange for processing your requests within a delayed window.
Trade latency for a discount
Batch endpoints run your requests asynchronously, often within a day, and price both input and output well below the real-time rate. The only cost is that you wait — no code rewrite of your prompts required.
What qualifies as batchable
Bulk classification, summarizing archives, generating content in advance, evaluations and back-office processing all tolerate delay. Anything a user is actively waiting on does not — keep those on the real-time API.
Calculators for this topic
Frequently asked questions
How much does the batch API save?+
Commonly around half the real-time token price, in exchange for delayed rather than instant results.
What work can I batch?+
Anything latency-tolerant — bulk classification, summarization, evaluations and pre-generated content.