OpenAI Decisions API: What It Is, vs Jev and Laya
What the OpenAI Decisions API does, its public-beta price and request format, and how it compares with Jev, Liquid AI d1 and the open-source Laya.
The OpenAI Decisions API is an OpenAI endpoint that answers typed questions instead of writing text: a yes/no probability, a pick from answers you define, or a rating on levels you define. You send the questions and some context (text or images), and it returns the answers with probabilities. OpenAI announced it at DevDay on September 29, 2026, opened it to all developers as a public beta on October 6, and runs it on GPT-6 Luna, its low-cost model.
That puts OpenAI in the same category as TypeSafe's Jev, Liquid AI's d1 and the open-source Laya: System One models, which make fast, bounded decisions while a larger model does the slow reasoning. This page summarises what OpenAI has published and how the OpenAI Decisions API compares with the other System One models.
OpenAI Decisions API at a glance
| What OpenAI has published | |
|---|---|
| Status | public beta since October 6, 2026; general availability expected "in the coming weeks" |
| Endpoint | POST /v1/decisions |
| Model | gpt-6-luna, the only model available |
| Question types | predicate (probability that a statement is true), choice (one of your values), score (expected level on your scale) |
| Output | choice and score answers include a probability per option and a confidence field |
| Input | a text string, or user messages with text and images; images must be inline base64 data URLs |
| Price | $0.10 per 1M input tokens; no charge for output, cache reads or cache writes |
| Speed | "about 10x faster than the Responses API"; at DevDay OpenAI showed 150 ms against 1.6 s |
| Data controls | Zero Data Retention and HIPAA for eligible customers; regional processing in the US and Europe |
Regional processing premiums and long-context multipliers apply to the price. OpenAI does not say how confidence is calculated, and the speed figures are OpenAI's own and have not been measured independently.
How the OpenAI Decisions API differs from Jev and Laya
| OpenAI Decisions API | TypeSafe Jev | Liquid AI d1 | Laya | |
|---|---|---|---|---|
| Who runs it | OpenAI (hosted) | TypeSafe (hosted) | Liquid AI (hosted), or you with the open d1-3B and d1-omni-600M | you, or our hosted Laya API |
| Weights | closed | closed | hosted d1 closed; d1-3B and d1-omni-600M open under the LFM Open License | open, Apache 2.0 |
| Input | text or images | text | text; d1-3B also images | text, including JSON |
| Question types | predicate, choice, score | choice, score, noul | choice, score, noul | choice, score, noul |
| Probabilities per option | yes, plus confidence | yes, calibrated | yes, calibrated | yes |
| API format | POST /v1/decisions, its own | POST /v1/systemone | POST /v1/systemone (TypeSafe SDK works) | POST /v1/systemone |
| Price | $0.10 per 1M input tokens | $0.042 per 1M input tokens | free tier; paid price not published | free to self-host |
| Speed | about 10x faster than the Responses API (OpenAI's figure) | 410 ms median in our test, network included | d1-3B: 8 ms on an RTX 4090 (Liquid's figure) | 11 to 54 ms locally in our test |
| Fine-tuning | not offered | no | not documented | yes, on your own labels |
The biggest practical differences:
- Images. The OpenAI Decisions API takes images as context, and so does Liquid's open d1-3B. Jev has accepted images since October 2026, as an experimental opt-in; Laya reads text only.
- A different wire format. Jev, d1 and Laya all speak
POST /v1/systemone, so code moves between them by changing the base URL. The OpenAI Decisions API uses its own endpoint and calls the yes/no typepredicateinstead ofnouland lists options as values and labelled levels, so switching to or from it needs a small adapter. - Price. Per 1M input tokens, OpenAI charges $0.10 and Jev $0.042, with output tokens free; Liquid has not published a paid price, and self-hosted Laya costs only your hardware.
- Control. Laya's weights are open: you can run it on your own servers, keep data in-house and fine-tune it. Liquid's open d1 models allow the same under their own license. The OpenAI Decisions API cannot be self-hosted, although OpenAI offers Zero Data Retention to eligible customers.
Which one should you use?
| Your situation | Start with |
|---|---|
| Decisions over screenshots, photos or other images | OpenAI Decisions API, or Liquid d1-3B if you want to run it yourself |
| You already use OpenAI and want one bill and one account | OpenAI Decisions API |
| Lowest hosted price per token | Clef-Flash on Workers AI ($0.038), then Jev or Microsoft-Decision-1 ($0.042) |
| Your stack is already on Azure | Microsoft-Decision-1, which takes the Jev request body at an Azure endpoint |
| You want code that works with several providers unchanged | the /v1/systemone format: Jev, d1 or Laya |
| Data must stay on your servers, or you want to fine-tune | Laya, self-hosted |
| You want open weights without running servers | our hosted Laya API |
If you are evaluating now, keep your questions in one place and run the same labelled sample through each backend. Converting a /v1/systemone request to the OpenAI format means renaming noul to predicate and rewriting the options as a list of values (choice) or labelled levels (score). Our System One benchmark shows how much accuracy differs between models on identical questions.
Is the OpenAI Decisions API a Jev competitor?
Yes, for hosted use. Both target the same job: a fast, cheap decision inside an agent or a pipeline, so the expensive model is only called when it is needed. OpenAI has not said the launch is a response to Jev. For teams already on OpenAI, the Decisions API removes the need for a second provider and adds image input; Jev is cheaper per token and shares a format with other providers, and Laya is the option when you need open weights, your own servers or fine-tuning.
FAQ
Is the OpenAI Decisions API available now?
Yes. It has been in public beta for all developers since October 6, 2026, and OpenAI expects general availability in the coming weeks. gpt-6-luna is the only model.
How much does the OpenAI Decisions API cost?
$0.10 per 1M input tokens. Output tokens, cache reads and cache writes are free. Regional processing premiums and long-context multipliers apply, and the beta price could change at general availability. For comparison, Jev costs $0.042 per 1M input tokens.
Does the OpenAI Decisions API return probabilities?
Yes. choice and score answers include a probability for each option and a confidence field, and predicate answers are a probability. OpenAI does not say how confidence is computed, so measure it on your own labelled sample before using it as a threshold.
Can I use the TypeSafe SDK or Jev code with the OpenAI Decisions API?
No. Jev, Liquid AI d1 and Laya share the /v1/systemone format; the OpenAI Decisions API has its own endpoint, POST /v1/decisions, with predicate in place of noul. You need OpenAI's SDK or a small adapter.
Is there an open-source alternative to the OpenAI Decisions API?
Yes. Laya is an open-weight System One model under Apache 2.0, with a 100+ language checkpoint. Liquid AI's d1-3B also reads images and has open weights under its own license. Other open models are listed on System One models compared and Jev alternatives.
Sources
- OpenAI Decisions API guide
- The Decoder: OpenAI expands Codex and its API at DevDay
- LiquidAI/d1-3B on Hugging Face
- TypeSafe: Introducing System One Models and Jev
Last verified: October 11, 2026.