> For the complete documentation index, see [llms.txt](https://docs.fastrouter.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.fastrouter.ai/explore-features/openai-decisions-and-typesafe-systemone/openai-decisions-gpt-6-luna.md). # OpenAI Decisions: GPT-6 Luna Get typed yes/no probabilities, choices, and rubric scores from OpenAI's gpt-6-luna through FastRouter's Decisions endpoint. *** ## Introduction FastRouter supports OpenAI's **Decisions API** through the `/decisions` endpoint. You send shared evidence (the **input**: text, images, or both) along with a list of **typed questions**, and get back structured answers: the probability that a condition is true, one option from a fixed set, or a score against a rubric. Use it to classify content, route requests, moderate images, triage issues, and prioritize work wherever your application needs a decision rather than generated text. **Why use Decisions through FastRouter** * **One API key:** use your existing FastRouter key, with no separate OpenAI account needed. * **Cost in every response:** `usage.cost` reports what each request cost. * **Observability:** every request appears in the Activity Log with tokens, latency, and cost. * **Switchable response format:** receive answers in the Decisions format (default) or the TypeSafe systemone format with `response_format`. > OpenAI's Decisions API is currently in **public beta**. *** ### Endpoint ```http POST https://api.fastrouter.ai/api/v1/decisions Authorization: Bearer Content-Type: application/json ``` ### Available models | Model | Description | | ------------ | -------------------------------------------------------------------- | | `gpt-6-luna` | OpenAI's decision model; currently the only model for this endpoint. | *** ### How a request works | Field | Type | Required | Description | | ----------------- | --------------- | -------- | ----------------------------------------------------------------------------------------------------------------------------------------------- | | `model` | string | Yes | `gpt-6-luna`. | | `input` | string \| array | Yes | Shared evidence for all questions: a text string, or user messages containing text and images. | | `questions` | array | Yes | The questions to evaluate. Each has a `type`, a `name`, `instructions`, and type-specific options. | | `response_format` | string | No | `"decisions"` (default) or `"systemone"`. See [Response formats](https://claude.ai/chat/1394a175-27d7-41bf-9c69-db45cb8377c5#response-formats). | Give every question a **unique `name`**. The API echoes it on each answer so you can match answers to questions. It is required if you use `response_format: "systemone"`, where answers are keyed by name. #### Question types | Type | Use it to | Main result | | ----------- | -------------------------------------------------------------------- | --------------------------------------------------------------- | | `predicate` | Check a condition, such as visible damage or whether text is urgent. | `probability`: estimate from 0 to 1 that the condition is true. | | `choice` | Select one unordered option, such as a department or category. | `choice`: one of the values you supplied. | | `score` | Rate against ordered levels, such as severity. | `score`: probability-weighted average of the level indices. | Use `choice` for categories with no order and `score` for ordered levels. Both return a probability distribution and a `confidence` value. *** ### Quickstart Route a customer complaint to a department. {% tabs %} {% tab title="cURL" %} ```bash curl https://api.fastrouter.ai/api/v1/decisions \ -H "Authorization: Bearer $FASTROUTER_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": "shipping", "description": "Delivery and tracking."}, {"value": "other", "description": "Requests outside these categories."} ] }] }' ``` {% endtab %} {% tab title="Python (requests)" %} ```python import os import requests resp = requests.post( "https://api.fastrouter.ai/api/v1/decisions", headers={"Authorization": f"Bearer {os.environ['FASTROUTER_API_KEY']}"}, json={ "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": "shipping", "description": "Delivery and tracking."}, {"value": "other", "description": "Requests outside these categories."}, ], } ], }, timeout=30, ) resp.raise_for_status() print(resp.json()["answers"][0]["choice"]) # "billing" ``` {% endtab %} {% endtabs %} #### Response ```json { "answers": [ { "type": "choice", "name": "department", "choice": "billing", "probabilities": [ { "value": "billing", "probability": 1 }, { "value": "technical", "probability": 0 }, { "value": "shipping", "probability": 0 }, { "value": "other", "probability": 0 } ], "confidence": 1 } ], "id": "fr_64320228-976f-40d3-91b6-1f81b3c5c10f", "model": "gpt-6-luna", "usage": { "input_tokens": 145, "output_tokens": 0, "total_tokens": 145, "cost": 0.0000145, "chat_id": "fr_64320228-976f-40d3-91b6-1f81b3c5c10f", "provider": "openai" } } ``` *** ### Question types in detail #### Predicate: check a condition Returns the probability that a condition is true. | Field | Type | Required | Description | | -------------- | ------------- | ----------- | ----------------------------------------------------- | | `type` | `"predicate"` | Yes | | | `name` | string | Recommended | Identifier echoed on the answer. | | `instructions` | string | Yes | The condition to check, phrased as a yes/no question. | **Example: check an image for visible damage** Images must be **inline base64 data URLs**. Hosted HTTP/HTTPS image URLs and `file_id` inputs are not supported. Combine `input_text` and `input_image` parts in a user message to evaluate an image together with context. **cURL** ```bash IMAGE_BASE64="$(base64 < product.png | tr -d '\r\n')" curl https://api.fastrouter.ai/api/v1/decisions \ -H "Authorization: Bearer $FASTROUTER_API_KEY" \ -H "Content-Type: application/json" \ --data-binary @- <Concept/decisions request/systemone requestContent to evaluateinputstateQuestion listquestions arrayquestions mapQuestion identifiernameMap keyYes/no questionpredicatenoulChoice optionschoices[] (value, description)criteria map (option → description)Score levelslevels[] (label, description)criteria array *** ### Response body
FieldTypeDescription
answersarray (or map with systemone)One answer per question.
idstringFastRouter request ID, also shown in the Activity Log.
modelstringModel that served the request.
usage.input_tokensintegerInput tokens billed.
usage.output_tokensintegerAlways 0; Decisions bills input tokens only.
usage.total_tokensintegerTotal tokens.
usage.costnumberCost of the request in USD.
usage.chat_idstringSame as id.
usage.providerstringUpstream provider (openai).
#### Refusals If the model declines to answer a question, that question's answer has `"type": "refusal"` and carries the question's `name`. Always check an answer's `type` before reading `probability`, `choice`, or `score`. ```json { "type": "refusal", "name": "visible_damage" } ``` *** ### Pricing `gpt-6-luna` on `/decisions` is billed on **input tokens only** at **$0.10 per 1M tokens**. There are no output, cache-read, or cache-write charges. Each response reports the exact charge in `usage.cost`. *** ### Interpreting answers * **Predicates** return the probability that the condition is true. * **Choice and score** return a full probability distribution plus `confidence`. * **Set thresholds on labeled examples** from your own application, weighing the cost of false positives against false negatives. Route low-confidence results to a human or a stronger model. ### Best practices * **Write questions around observable criteria.** "Is there a visible crack?" is better than "Is this product bad?" * **One concern per question.** Split compound checks into separate questions in the same request. * **Make options distinct.** Choices should not overlap, and adjacent score levels should have clearly different criteria. * **Batch independent questions** about the same input in one request; it's cheaper and faster than separate calls. * **Use Structured Outputs instead** when you need a generated object (extracted fields, written explanations) rather than a probability, choice, or score. ### Common use cases * **Support triage:** urgency (predicate), department (choice), severity (score). * **Image moderation and QA:** damage, policy violations, or product category from photos. * **Content moderation:** policy checks with calibrated probabilities. * **LLM output grading:** score other models' responses for accuracy or tone. * **Agent guardrails:** check a proposed action before allowing it. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://docs.fastrouter.ai/explore-features/openai-decisions-and-typesafe-systemone/openai-decisions-gpt-6-luna.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.