OpenAI Decisions API Hits Public Beta With 10x Faster Typed Answers

OpenAI has put its Decisions API into public beta, MarkTechPost reports. It takes text, images or both and returns typed answers code can act on: a probability for a yes-or-no check, a pick from a list, or a score on an ordered scale. OpenAI says it runs about 10x faster than its Responses API. The only supported model today is gpt-6-luna.

Pricing is simple. Input costs $0.10 per 1M tokens, with no charge for output or caching, though regional premiums and long-context multipliers still apply. It supports Zero Data Retention and HIPAA for eligible customers, with US and European data residency.

This targets one of the most common patterns in AI apps: ask a model a question, then scrape a label out of its reply. Making that a first-class product is a meaningful shift.

We think this is a sensible, overdue product. Plenty of real AI work is classification: is this photo damaged, which queue does this ticket go to, how severe is this outage. Asking a chat model to write prose and then parsing it is wasteful and fragile. Returning a probability directly is cheaper, faster and easier to audit.

The probability part is the bit we like most. A score with a number attached forces teams to decide what threshold triggers an action. That is healthier than a confident sentence nobody questions. OpenAI's docs even advise setting thresholds with labeled examples from your own application, which is the right instinct.

But here is the gap. OpenAI has not published accuracy or calibration data for the endpoint. A probability is only useful if 70 percent really means right about 70 percent of the time. Until independent tests show that, the numbers are a promise, not a guarantee. The speed figure is also OpenAI's own claim, and coverage put a decision near 150 milliseconds against about 1.6 seconds for regular Luna calls.

The competitive angle is hard to miss. TypeSafe's Jev, launched in September, does the same kind of thing and is cheaper on input at $0.042 per 1M tokens. OpenAI's edge is image input, compliance options and an open beta anyone can try. This looks like a platform responding fast to a startup that found a real gap.

The downside if this scales is the familiar one. Cheap, instant decisions make it easy to automate judgments that used to get a human glance, in moderation, claims and screening. We would like to see OpenAI ship calibration data and guidance on human review before general availability.

OpenAI positions Decisions for probabilities, choices and scores, Structured Outputs for filling a JSON schema or writing explanations, and function calling for tool requests. It says general availability is expected in the coming weeks.