When to use the SDK

2026-08-11By Aditya Perswal

There are a lot of pipelines out there doing LLM enrichment. Checking your catalog descriptions for fraudulent information. Summarizing the chat history from your customer service texts. Taking a pile of notes on a lead and cataloguing them into your CRM. None of them need to be instant. They can all wait a few minutes to drop costs by 48%.

Those pipelines, running thousands of jobs a minute, can use us really easily. To use us, you change the API URL, you change the API key, and you add one parameter. start_within says how long you're willing to wait, from five seconds up to ten minutes. If the cheaper tier starts inside your window, you pay about half. If it doesn't, your request runs on the standard tier at full price and the pipeline stays on schedule.

client.responses.create(
    model="gpt-5.6-sol",
    input=prompt,
    extra_body={"start_within": "00h-05m-00s"},
)

Background work in your app

When you're building a software application that does work in the background, we're great for that too. Browser agents, deep research, building up a user profile. Put the non-latency-sensitive work through us with a deadline, and keep the latency-sensitive calls as they are, because we support default and priority requests on the same key.

Side projects that don't waste tokens

When you want to build side projects without wasting tokens, we're great for that as well. You're messing around, you're doing some other work at the same time, and you're willing to wait five seconds. Set a deadline on every call and most of what the project sends costs half.