OpenAI Decisions API

OpenAI announced the Decisions API at DevDay on September 29, 2026, in limited preview.^[1] It targets the same job as TypeSafe's Jev: return a decision your code can act on, rather than text it has to parse.

What it is

The Decisions API turns a classification or routing step into a single call. A developer supplies context — text or an image — together with a set of pre-defined answers. The endpoint returns the answer that fits best, instead of a paragraph that has to be parsed.^[1]

Because the answer space is finite, the model cannot answer outside the list. OpenAI positions it for classification, request routing, and choosing an agent's next action.^[1]

What OpenAI has actually confirmed

Confirmed

• The API was announced at DevDay on September 29, 2026.^[1]
• It is in limited preview, with a broader rollout described as coming in the following days.^[1]
• It focuses Luna's intelligence on user-defined questions with finite pre-defined answers — in other words, it runs on Luna.^[1]
• It accepts context as text or images and returns an answer for classification, routing, or an agent's next action.^[1]

Not yet confirmed

• Pricing — OpenAI had not published a price at the time of writing.
• Latency — no official figure has been published.
• Accuracy or benchmark figures — none formally published.
• How the model is applied under the hood beyond running on Luna.

How it works

The developer defines a question and the set of allowed answers. The endpoint scores only those options and returns the chosen one. Skipping free-text generation is what keeps the output constrained to the list the developer provided.^[1]

Availability & pricing

Availability: limited preview as of September 29, 2026, with a broad release planned in the coming days.^[1]

Pricing: Not yet confirmed. OpenAI had not published a price for the Decisions API at the time of writing.

How it compares

The Decisions API answers a question by choosing from predefined options, which is the same core task Jev by TypeSafe AI was built for. Jev returns typed primitives — choice, score, or yes/no — with calibrated probabilities,^[2] while the open-source Laya does the same self-hosted.^[3] Read the full Decisions API vs Jev comparison for the point-by-point breakdown.