OpenAI Decisions API: what it is and how it compares to Jev

What OpenAI announced

In its DevDay 2026 recap, OpenAI describes the Decisions API as real-time decision-making that “focuses Luna’s intelligence on a specific set of user-defined questions with finite pre-defined answers.” You supply context as text or images and get back an answer you can use to classify content, route requests, or pick an agent’s next action.

Launch status: limited preview, with a broad release planned “in the coming days.” OpenAI also launched computer use in the Agents API at the same event, and third-party coverage frames the Decisions API as the fast routing layer for those agents.

• Announced: 29 September 2026, OpenAI DevDay
• Model family: Luna (OpenAI’s wording)
• Input: text or image context, plus questions with a fixed answer set
• Output: one of your answers. It cannot invent a new label.
• Status: limited preview. No public reference or pricing yet.

Reported speed and accuracy

The numbers so far come from launch coverage, not an OpenAI benchmark page. The Decoder reports a computer-task simulation with 76 correct choices out of 78 scored steps at about 230 ms per call. byteiota puts typical latency near 150 ms and says it is roughly 10× faster than calling standard Luna through chat.

Treat these as early claims. Measure p95 latency and accuracy on your own questions once you have access. The same advice applies to Jev’s vendor charts.

OpenAI Decisions API vs Jev

Both products do the same job: take state plus a question with fixed answers, and return a constrained value instead of generated prose. The differences are in the inputs, the price, and how mature each one is.

• Input: OpenAI Decisions accepts text or images. Jev is text-only.
• Question types: Jev has choice, score (ordinal), and noul (yes/no probability), with up to 255 options and 64k context. OpenAI has not published its question schema.
• Price: Jev is $0.042 / 1M input tokens with free output. OpenAI pricing is not announced.
• Availability: Jev’s API is open, and this site has a live playground and keys. OpenAI Decisions is limited preview only.
• Platform: OpenAI Decisions sits next to the Agents API and computer use. Jev runs on TypeSafe, OpenRouter, Vercel AI Gateway, and jevmodel.org.

Which one to pick

If your decision depends on a screenshot, a camera frame, or a rendered UI, only OpenAI Decisions accepts images. That is a real advantage once it reaches general availability, especially if your agents already run on OpenAI.

For high-volume, text-only routing, triage, and guardrails, Jev is shipping today at a published price with a typed response contract. Keep your questions in code so you can run the same set against both APIs and compare accuracy, calibration, and cost once OpenAI opens access.

Try the same pattern with Jev now

This is the shape you would port: state, named questions, and a fixed set of answers. It runs on jevmodel.org today with a Dashboard key.

curl -X POST https://jevmodel.org/v1/systemone \
  -H "Authorization: Bearer $JEVMODEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "Agent finished step 3: invoice PDF parsed, total mismatch detected.",
    "questions": {
      "next_agent": { "type": "choice", "instructions": "Which agent handles the next step?",
        "criteria": { "billing": "Payment or invoice issue", "support": "Needs a human reply", "done": "Nothing left to do" } }
    }
  }'