If I run batches consisting of 1 JSONL line, will it still be 50% of the cost?

API
1

If I run many batches consisting of of 1 JSONL line, will it still be 50% of the cost?

I don’t quite understand why uploading batches is recommended to be by multi-line JSONL file rather that creating a new batch for each request.

Which advantages may I have by submitting long batches rather than short ones?

2

Hey, @porton.victor. Hope you’re well!

Do single-line batches still get the 50% discount from OpenAI?

Yes! Even if your batch consists of just one JSONL line, you still get the 50% discount. OpenAI applies this discount to all work done through their Batch API endpoint, regardless of batch size.

So, why group multiple requests into bigger batches?

While you don’t get extra discounts beyond the initial 50%, running larger batches is usually better in terms of operational efficiency.

  1. Less Overhead:

  2. Better Throughput:

  3. Simpler Coordination:

Downsides of Many Small (1-line) Batches:

  1. Higher Operational Complexity:

  2. Slower Job Management:

So…You always get the 50% discount on batch API usage regardless of size. Using larger batches primarily helps with operational ease, speed, and resource management, rather than further cost savings.

Hope this helps!

3

If using o4-mini or o3 AI models, you can also use the API parameter "service_tier": "flex", for lower-priority processing that is not as extreme as “24-hour turnaround” for a similar discount. You can run as a “background” call if using Responses, to then poll and “pick up” your response ID when done.

The performance of other models like gpt-4.1, gpt-4o, and gpt-5 has been greatly affected recently, and yet this symptom is resolved by using "service_tier": "priority" for higher expense, so these other models got stealth-inflated in cost anyway.