OpenAI just took its Decisions API out of beta. If you build agents or automations, this one is worth a look, because a lot of what our bots do all day isn’t writing paragraphs. It’s deciding stuff. Is this spam? Which queue does this ticket go to? How risky is this request?
What it does
- You send input (text, images or JSON) plus a list of questions to
POST /v1/decisions. - Each question comes back typed: a probability that something is true, one choice from a list you define, or a score.
- It only runs on
gpt-6-luna, and OpenAI says it can be up to 10x faster than asking Luna the same thing through the Responses API. - GA bumps the limit from 200 to 400 questions per request, and adds prompt caching and native GPT-Live integration.
- Pricing: about $0.10 per million uncached input tokens, $0.01 cached, and no charge for output tokens. Cache writes are billed separately.
My take
This is the right shape for a lot of agent plumbing. Routing and classifying with a full chat completion is like using a sledgehammer to hang a picture. A typed answer is easier to test, easier to log and way harder for a model to ramble its way out of. I’d still keep a human or a hard rule in front of anything that sends, buys or deletes. A fast decision is still just a decision, and it can be wrong. Azure OpenAI has a version of it too, if that’s where your stuff lives.
Have a blessed day! And Happy Hacking!
