# GPT-6 Luna Decisions - Core.Today AI API

> OpenAI's Decisions API on GPT-6 Luna, through the gateway. Not chat: send an input (text, or user messages with text and base64 images) plus typed questions - predicate (probability a condition is true), choice (one of your options with per-option probabilities) or score (a position on your ordered levels) - and get one typed answer per question with confidence. Roughly 10x faster than a Responses call and billed on input tokens only.

- **Provider**: OpenAI
- **Model ID**: gpt-6-luna-decisions
- **Category**: LLM
- **Credits**: 0.19 per 1K input tokens (output free)
- **Speed**: Fast
- **Quality**: High

## Model Specifications
- **Context Window**: 1.1M tokens
- **Max Output**: 0K tokens
- **Training Cutoff**: 2026-05-18
- **Supported Formats**: json
- **Compatible SDK**: OpenAI, HTTP

### Capabilities
- Vision (image input)
- JSON Mode

### Token Pricing (per 1M tokens)
- **Input Tokens**: 185.8 credits ($0.12)
- **Output Tokens**: not charged — billed on input tokens only

## Features
- Three question types: predicate (0-1 probability), choice (your values + probabilities + confidence), score (weighted position on ordered levels)
- Text and image input (input_text / input_image parts, base64 data URLs only)
- Many questions per request over one input - independent answers, one input-token charge
- Input-token billing only; output, cached and cache-write tokens are free
- A question can come back as a refusal; the other answers are still returned
- OpenAI SDK support: client.decisions.create(...) with base_url https://api.core.today/llm/openai/v1

## Use Cases
- Content classification and request routing with calibrated probabilities
- Visual checks on product photos (damage, category) with predicate questions
- Severity, priority and quality scoring against your own rubric
- Guardrails and relevance filtering before a larger model runs

## API Endpoint

Base URL: https://api.core.today
Endpoint: POST /llm/openai/v1/decisions

## Authentication
Header: Authorization: Bearer YOUR_API_KEY

Note: LLM endpoints use OpenAI-compatible format with Authorization Bearer token.

## Input Parameters

### Required
- **model**: string (default: gpt-6-luna-decisions) - gpt-6-luna-decisions (catalog id) or gpt-6-luna - the gateway bills the Decisions row either way and sends gpt-6-luna upstream.
- **input**: string | message[] - The evidence shared by every question: a text string, or user messages whose content holds input_text and input_image parts (images as data:image/...;base64 URLs).
- **questions**: array - Array of questions, each with a unique name, a type and instructions. choice adds choices [{value, description}], score adds ordered levels [{label, description}] from lowest to highest. Answers come back under the same names.

## Examples

### Route a complaint and rate its severity
A predicate, a choice and a score in one call
```json
{
  "model": "gpt-6-luna-decisions",
  "input": "I was charged twice for my order and support has not replied for 3 days.",
  "questions": [
    {
      "type": "predicate",
      "name": "is_urgent",
      "instructions": "Does this message convey urgency?"
    },
    {
      "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": "Requests outside these categories."
        }
      ]
    },
    {
      "type": "score",
      "name": "severity",
      "instructions": "How severe is this issue for the customer?",
      "levels": [
        {
          "label": "Minor",
          "description": "Inconvenience only."
        },
        {
          "label": "Significant",
          "description": "Money or access affected."
        },
        {
          "label": "Critical",
          "description": "Blocked with no workaround."
        }
      ]
    }
  ]
}
```

### Check a product photo for damage
Image input as a base64 data URL plus a predicate question
```json
{
  "model": "gpt-6-luna-decisions",
  "input": [
    {
      "role": "user",
      "content": [
        {
          "type": "input_text",
          "text": "Inspect the product in this photo."
        },
        {
          "type": "input_image",
          "image_url": "data:image/png;base64,<BASE64>"
        }
      ]
    }
  ],
  "questions": [
    {
      "type": "predicate",
      "name": "visible_damage",
      "instructions": "Does the product have visible damage, such as a crack, tear, or dent? Ignore shadows and damage to the packaging."
    }
  ]
}
```

## Response Format
```json
{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Response text here"
      },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 100,
    "completion_tokens": 50,
    "total_tokens": 150
  }
}
```

## Tips
- Read answers by type: predicate.probability (0-1), choice.choice + probabilities[] + confidence, score.score (can fall between levels) + probabilities[] + confidence; handle type refusal
- Give every question a unique name and put the full question in instructions; add an 'other' choice when your list may not cover every input
- Put every question about the same input in one request - answers are independent and the input is charged once
- Hosted image URLs and file ids are not accepted - inline the image as a base64 data URL
- stream is not supported on this endpoint through the gateway; the chat gpt-6-luna row is unaffected and keeps its own pricing

## Documentation
https://developers.openai.com/api/docs/guides/decisions