OpenAI Decisions API · EU alternative

Typed decisions, processed in the EU

The OpenAI Decisions API (generally available since 22 October 2026) answers yes/no, choice and score questions with gpt-6-luna. Decision Models offers the same three question types on open-weight models, with an EU tier, a public DPA and rates from €0.018 ($0.021) per 1M input tokens. To be fair: OpenAI offers processing in Europe too. Facts checked 25 October 2026.

Get API key See the EU tier

At a glance

OpenAI Decisions API and Decision Models compared

OpenAI facts from OpenAI’s Decisions guide, retrieved 25 October 2026. Decision Models rates exclude taxes.

Feature	OpenAI Decisions API	Decision Models
Input price per 1M tokens	$0.10 uncached; cache hits are discounted, cache writes are charged; regional processing premiums and long-context multipliers apply	s1-fast from €0.018 ($0.021) in the EU tier, €0.017 ($0.019) Peer-to-Peer; s1-pro €0.074 ($0.084) EU. The EU rate is the list rate; no further regional premium. All rates
Output tokens	Not charged	Not charged
Models	`gpt-6-luna` only (proprietary)	Several open-weight models (Apache-2.0); every answer names the model version
Questions per request	1–400	s1-pro up to 3, s1-fast and s1-vision 1. More questions means more requests.
Input types	Text, images, JSON, conversations with tool calls	Text or JSON as `state`; one image with s1-vision
Prompt caching	Yes	No
Processing in Europe	Yes: regional processing in Europe (EEA + Switzerland) and data residency; eligibility per OpenAI’s data controls	EU tier: gateway and GPU inference only at contracted operators in the EU; EU requests never leave the EU. Where inference runs
Retention	A no-retention option for eligible customers (per OpenAI)	Request and response content is not stored on our gateway or inference servers
DPA	Per OpenAI’s data controls	Public DPA, accepted at signup
Provider	OpenAI	Decision Models, a company based in Passau, Germany (imprint)

Migration

From /v1/decisions to /v1/systemone

The three question types map one to one. Field names differ; the table shows every mapping.

Field	OpenAI Decisions API	Decision Models
Endpoint	`POST https://api.openai.com/v1/decisions`	`POST https://api.decisionmodels.io/v1/systemone`
Key	`Authorization: Bearer $OPENAI_API_KEY`	`Authorization: Bearer $DECISIONMODELS_API_KEY` (from the dashboard)
Model	`"model": "gpt-6-luna"`	`"model": "s1-fast"`, `"s1-pro"` or another id from `GET /v1/models`
Shared input	`input`: a string or an array of messages (also tool calls, `input_json`)	`state`: a string or a JSON object/array. Flatten a conversation into one `state` first.
Questions	`questions`: an array, each question has a `name`	`questions`: an object keyed by the question name
Yes/no	`"type": "predicate"` → answer `probability`	`"type": "noul"` → answer `noul` (probability of “yes”)
Choice	`"type": "choice"` with `choices: [{"value", "description"}]`	`"type": "choice"` with `criteria: {"value": "description"}` (at least two options)
Score	`"type": "score"` with 2–10 `levels: [{"label", "description"}]`	`"type": "score"` with an ordered `criteria` array (at least two levels); `score` is also the probability-weighted average of the level indices, starting at 0
Answers	`answers`: an array in question order	`answers`: an object keyed by question name, plus `model_version` and `usage`
Region	Data residency or regional processing per project (United States or Europe)	Header `S1-Region: eu\|global`, or the key’s default tier
Images	`input_image` parts in `input` or in questions	`s1-vision` via `/v1/multimodal`, exactly one image per request

Example

The same question, before and after

A complaint is routed to a department. The option descriptions move from choices to criteria.

Before · OpenAI

curl https://api.openai.com/v1/decisions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-6-luna",
       "input":"I was charged twice for my order.",
       "questions":[{"type":"choice","name":"department",
         "instructions":"Which department should handle this complaint?",
         "choices":[{"value":"billing","description":"Payments, invoices, and refunds."},
                    {"value":"technical","description":"Problems using the product."},
                    {"value":"other","description":"Anything else."}]}]}'

After · Decision Models (EU)

curl https://api.decisionmodels.io/v1/systemone \
  -H "Authorization: Bearer $DECISIONMODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -H "S1-Region: eu" \
  -d '{"model":"s1-fast",
       "state":"I was charged twice for my order.",
       "questions":{"department":{"type":"choice",
         "instructions":"Which department should handle this complaint?",
         "criteria":{"billing":"Payments, invoices, and refunds.",
                     "technical":"Problems using the product.",
                     "other":"Anything else."}}}}'

Honest fit

When OpenAI is the better fit

You need many questions in one request (up to 400), prompt caching, conversations with tool calls, or images inside questions.
You already have an OpenAI contract with European data residency.

Decision Models fits when you want open-weight models, low per-token rates, a public DPA and a provider based in Germany. Test the answers on your own labelled examples before you switch production traffic.

OpenAI and gpt-6-luna are trademarks of OpenAI. Decision Models is not affiliated with OpenAI and uses the names only for comparison.