OpenAI-Compatible Decisions API

Typed decisions shipped on /v1/systemone, which speaks TypeSafe's request format. Now that OpenAI has its own Decisions API, a second wire format per provider would mean rewriting your integration every time you switch models. /v1/decisions is a new OpenAI-compatible endpoint that serves every decision model with one request shape, and the OpenAI SDKs work against it by changing the base URL.

GPT-6 Luna Decisions, with image input

gpt-6-luna-decisions is GPT-6 Luna served through OpenAI's Decisions API. It reads text or images and answers predicate, choice and score questions with calibrated probabilities.

Output, cache reads and cache writes are free on all three — the response is a set of values, not generated text. gpt-6-luna-decisions has a 1.05M-token context and is in beta, matching OpenAI's public beta of the API.

1curl -X POST "https://api.llmgateway.io/v1/decisions" \2  -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "gpt-6-luna-decisions",6    "input": "I was charged twice for my order.",7    "questions": [8      {9        "type": "choice",10        "name": "department",11        "instructions": "Which department should handle this complaint?",12        "choices": [13          { "value": "billing", "description": "Payments, invoices, refunds" },14          { "value": "technical", "description": "Problems using the product" }15        ]16      },17      {18        "type": "predicate",19        "name": "is_urgent",20        "instructions": "Does this convey urgency?"21      }22    ]23  }'

1curl -X POST "https://api.llmgateway.io/v1/decisions" \2  -H "Authorization: Bearer $LLM_GATEWAY_API_KEY" \3  -H "Content-Type: application/json" \4  -d '{5    "model": "gpt-6-luna-decisions",6    "input": "I was charged twice for my order.",7    "questions": [8      {9        "type": "choice",10        "name": "department",11        "instructions": "Which department should handle this complaint?",12        "choices": [13          { "value": "billing", "description": "Payments, invoices, refunds" },14          { "value": "technical", "description": "Problems using the product" }15        ]16      },17      {18        "type": "predicate",19        "name": "is_urgent",20        "instructions": "Does this convey urgency?"21      }22    ]23  }'

Answers come back as an array in question order, each echoing its name. A question the model declines returns refusal while the rest are still answered. To evaluate an image, send a user message with input_text and input_image parts.

Microsoft-Decision-1 on Azure

microsoft-decision-1 is Microsoft's small decision model for classification, routing, ranking and grading, with a 32K-token context. Call it as microsoft-decision-1 with LLM Gateway credits, or bring your own Azure key: deploy it in Microsoft Foundry under the deployment name Decision-1 and add your Azure key and resource. It takes text only.

Jev on the same endpoint

jev-1.13.0 works on /v1/decisions too, at the same price. The gateway translates the request to TypeSafe's format and the answers back, so switching between Jev, Microsoft-Decision-1 and GPT-6 Luna Decisions is a one-field change. Jev is text-only: image parts are rejected with a 400.

/v1/systemone will be deprecated soon

/v1/systemone keeps working for existing integrations, but it now serves TypeSafe models only and gets no new models. Requesting gpt-6-luna-decisions or microsoft-decision-1 there returns a 400 pointing at /v1/decisions. We will announce a deprecation date separately; the migration guide maps every field.

Both endpoints run the same machinery as the rest of the gateway: IAM rules, compliance policies, credit gating, provider key rotation, request logging with a per-model cost breakdown, and a per-organization limit of 600 requests per minute.

Decisions docs → | Browse decision models →