GPT Image 2.5 Flare & Sunburst are live on EvoLinkTry GPT Image 2.5
EvoLink.AIEvoLink.AI

GPT-6 Luna Decisions API

Access OpenAI GPT-6 Luna Decisions—also searched as OpenAI Decisions API / gpt-6-luna-decisions— through EvoLink's unified API on the Decisions endpoint. Test typed answers, image input before integrating.

OpenAIText GenerationAvailable
from $0.100 / 1M input tokens
API docs
Decisions APIPredicate questionsChoice questionsScore questionsImage inputInput-only billing
Production routeLive
Context
1.05M context
Best For
Classification, routing, moderation checks, rubric scoring
Input
Text + images
Output
Typed answers: probability · choice · score
ModelsPricingReadmeAPI

Choose GPT-6 Luna Decisions

One model ID on the Decisions endpoint. The card lists the input rate below and above the long-context threshold; output and cache tokens are not billed.

GPT-6 Luna Decisions

GPT-6 Luna on OpenAI's Decisions API for typed answers at volume.

Selected
Model ID
gpt-6-luna-decisions
Best for

Classification, routing, moderation checks, and rubric scoring where you need a probability or a fixed option instead of generated text

< 272K
Input
$0.100 / 1M
6.8 cr / 1M
Output
Not billed
≥ 272K
Input
$0.200 / 1M
13.6 cr / 1M
Output
Not billed

All rates are per 1M tokens, shown in USD and credits, and reflect your account's current pricing. Above 272K prompt tokens the long-context tier applies (×2) to every token role. On this endpoint only input tokens are billed: output, cache-read, and cache-write tokens are free.

GPT-6 Luna Decisions pricing

Estimate what one GPT-6 Luna Decisions request costs before you integrate. The calculator uses your account's current rates, with official pricing as a reference.

Request calculator

Enter the token mix for one request.

Estimated request cost

GPT-6 Luna Decisions
Short-context rate
USD$0.0002
Credits0.01
Input tokens0.0034 cr

Minimum charge: 0.01 credits per request. Above 272K input tokens the whole request is billed at ×2.

Budget guide

Approximate requests using the current token mix.
Add credits
$10
About 65000 requests

For quick testing

$50
About 340000 requests

For regular development

$100
About 680000 requests

For production evaluation

GPT-6 Luna Decisions API

Classify support tickets, route requests, and score text or images with GPT-6 Luna on OpenAI's Decisions API. Get a probability, a fixed choice, or a rubric score in one synchronous call through EvoLink's unified gateway, using the same API key and input-only billing. The Decisions API is in public beta.

GPT-6 Luna Decisions is served on EvoLink under the model ID gpt-6-luna-decisions through Decisions API, with the same API key and balance you use for every other model. It offers a 1.05M context window, plus typed answers, image input.

GPT-6 Luna Decisions

GPT-6 Luna Decisions specs and capabilities

Numbers come from the EvoLink route configuration; capabilities are what the API exposes today.

Context window
1.05M tokens
Input
Text + images
Output
Typed answers: probability · choice · score
Long-context tier
×2 above 272K prompt tokens
Protocols
Decisions API
Model ID
gpt-6-luna-decisions

GPT-6 Luna Decisions for classification, routing, and scoring

Use Decisions when application logic needs a probability, a fixed choice, or a rubric score. Keep generated replies, field extraction, and tool calls on GPT-6 Luna through Chat Completions or Responses; both routes remain accessible through the same EvoLink gateway.

Flag Tickets or Photos for Review

Use a predicate question to check whether a ticket needs escalation, a passage is relevant, or a photo shows damage. Read the probability that the condition is true and choose a review threshold from your labeled examples.

Classify Support Tickets and Route Requests

Define departments such as billing, technical support, and account access with a short description for each. A choice question returns the selected value and each option's probability so your application can decide where to send the request.

Prioritize Work with Rubric Scores

Define ordered severity levels such as cosmetic, workaround available, and fully blocked. A score question returns a probability-weighted average that can fall between levels, helping your application prioritize work against the same rubric.

Assess Product Photos with Their Descriptions

Combine a customer's description with inline base64 product photos in one user message. Ask independent questions about visible damage, category, or severity against the same evidence in a single request.

Two ways to use GPT-6 Luna Decisions: EvoLink API or Agent

Use the EvoLink API for product backends and batch jobs, or call GPT-6 Luna Decisions from Codex, Claude, or Gemini to label, route, and score items inside a workflow. Both paths share the same EvoLink API key, balance, model ID, and request history.

Option 1

Integrate with the EvoLink API

Best for: product backends, batch jobs, automated pipelines

Send OpenAI-compatible Decisions requests to EvoLink: the evidence goes in input, and each question declares its type, instructions, and allowed choices or score levels.

  1. 1Create an EvoLink API key in the console
  2. 2Point the OpenAI SDK at the EvoLink base URL, or POST to /v1/decisions, with the model ID shown above
  3. 3Send one representative request and read usage.input_tokens; it is the only billed quantity
  4. 4Give every question a unique name and set your own thresholds on the returned probabilities
Create EvoLink API Key
Option 2

Call it with an Agent

Best for: labeling, routing, and scoring steps in Codex, Claude, and Gemini

Give the Agent the evidence, the questions, and the acceptance criteria. It assembles the request, calls GPT-6 Luna Decisions through EvoLink, and returns the typed answers with token usage.

  1. 1Set EVOLINK_API_KEY in your local environment; never put it in code or a prompt
  2. 2Describe the evidence to evaluate and the questions to ask, with choices or score levels
  3. 3Ask the Agent to call GPT-6 Luna Decisions through EvoLink and show the request before sending
  4. 4Let the Agent report each answer, the input token usage, and any error body

GPT-6 Luna Decisions API code example and error handling

This example shows the shortest runnable request: one piece of text evidence and one choice question sent to the Decisions endpoint. Open the API tab for the complete parameter and response reference.

View complete API docs
cURL
curl -X POST https://api.evolink.ai/v1/decisions \
  -H "Authorization: Bearer $EVOLINK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-luna-decisions",
    "input": "My invoice shows two charges for the same month.",
    "questions": [
      {
        "type": "choice",
        "name": "ticket_type",
        "instructions": "Which category best describes this support ticket?",
        "choices": [
          { "value": "billing", "description": "Charges, invoices, refunds" },
          { "value": "bug", "description": "Something is broken" },
          { "value": "how_to", "description": "How to use a feature" }
        ]
      }
    ]
  }'

# The response has one answer per question, matched by name:
# answers[0].choice is the selected value and answers[0].probabilities
# lists a probability for every option. Only usage.input_tokens is billed.
# The endpoint is synchronous and rejects fields it does not define, such
# as stream, temperature, and max_tokens.

Unknown or missing parameter

The endpoint rejects fields it does not define. Remove stream, temperature, max_tokens, and other Chat Completions parameters, and check that input and questions are present.

Image is not a base64 data URL

Send images inline as data:image/...;base64 URLs. Hosted image URLs and file IDs are rejected.

Authentication or balance issue

Check the Authorization bearer token and confirm the available balance in the console.

Duplicate question names or a non-user role

Question names must be unique within one request, and every input message must use the user role.

Rate limited (429)

Back off and retry with jitter; batch or queue requests instead of sending parallel bursts.

Before you take GPT-6 Luna Decisions to production

Four checks that decide cost and reliability.

01

Send only the documented fields

The endpoint rejects unknown fields. Remove stream, temperature, max_tokens, and other Chat Completions parameters before you send.

Required
02

Inline images as base64 data URLs

Hosted image URLs and file IDs are not accepted, and one request takes at most 128 image parts.

Required
03

Name every question and set thresholds from labeled examples

Names must be unique within a request and are echoed in the answers; choose thresholds from the cost of false positives and false negatives in your own data.

Recommended
04

Budget from usage.input_tokens

Even a one-line request counts roughly 150 input tokens of fixed overhead; a minimum charge applies per request, and above 272K input tokens the long-context rate applies.

Recommended

Input-only billing

Only input tokens are billed, including image tokens. Output, cache-read, and cache-write tokens are free on this endpoint, and each request has a minimum charge.

Long-context threshold

At more than 272K input tokens the whole request is billed at 2x the input rate.

GPT-6 Luna Decisions vs GPT-6 Luna on Chat Completions

EvoLink

Compare the Decisions endpoint with calling GPT-6 Luna through Chat Completions or Responses, using EvoLink prices per 1M tokens.

GPT-6 Luna Decisions
EvoLink price / 1M (input / output)$0.1 / $0
Context window1.05M
ResponseTyped answers: probability, choice, score
Best forClassification, routing, moderation checks, and rubric scoring where you need a probability or a fixed option instead of generated text
GPT-6 Luna
EvoLink price / 1M (input / output)$0.092 / $0.45
Context window1.05M
ResponseGenerated text, JSON, tool calls
Best forTasks that need generated text, fields in your own schema, or tool calls

Prices reflect your account's current pricing; signed-out visitors see Default rates. The Decisions endpoint bills input tokens only, so its output price is zero.

GPT Model Family

Compare all GPT models
GPT-6 Luna

GPT-6 Luna

The same model on Chat Completions and Responses for generated text, JSON, and tool calls

View model
GPT-6.1 Sol

GPT-6.1 Sol

Updated Sol for multi-file coding and Responses-based agents

View model
GPT-6 Astra

GPT-6 Astra

OpenAI's most capable model for the hardest end-to-end work

View model

Other text models on EvoLink besides GPT-6 Luna Decisions

Gemini 3.5 Flash-Lite

Gemini 3.5 Flash-Lite

Google's lowest-cost Flash Lite model with multimodal input, built for high-volume, latency-sensitive, and batch workloads.

View model
DeepSeek V4.1 Flash

DeepSeek V4.1 Flash

DeepSeek Flash route with text and image input, a 1M context window, and configurable thinking.

View model
GLM-5.3 Flash

GLM-5.3 Flash

Z.ai's multimodal volume tier of the GLM-5.3 generation, with a 1M-token context window.

View model
Grok 4.5

Grok 4.5

xAI reasoning and tool-use model with a 500K context window and metered server-side tools.

View model

GPT-6 Luna Decisions API - FAQ

Which model ID should I use on the Decisions endpoint?

Use gpt-6-luna-decisions with POST /v1/decisions. The OpenAI model ID gpt-6-luna is also accepted on this endpoint, so code written for the OpenAI SDK works after you change the base URL. Usage and billing records list these requests as gpt-6-luna-decisions.

How much does GPT-6 Luna Decisions cost on EvoLink?

The default rate is $0.100 per 1M input tokens. Output, cache-read, and cache-write tokens are not billed. Each request has a minimum charge, and requests above 272,000 input tokens are billed at 2x the input rate. Logged-in users see the rate for their own account.

What question types are supported?

Three: predicate returns the probability that a condition is true; choice returns one of the values you supply, with a probability for each option; score rates the input against ordered levels and returns a probability-weighted average. A request can mix types, and each question is answered independently.

Can I send images?

Yes. In a user message, add input_image parts with inline base64 data URLs next to input_text parts. Hosted image URLs and file IDs are rejected, and one request accepts at most 128 image parts. Image tokens are billed as input.

Does it support streaming, system prompts, or tools?

No. The Decisions endpoint is synchronous and accepts only model, input, questions, and an optional safety_identifier. Messages must use the user role, and fields such as stream, temperature, and max_tokens are rejected with a 400 error.

When should I use GPT-6 Luna on Chat Completions or Responses instead?

Use Decisions to classify a ticket, check a condition, or score an input when your application needs a probability, a fixed option, or a score. Keep GPT-6 Luna on Chat Completions or Responses when you need a written reply, extracted fields in your own JSON schema, or tool calls. Both are available through the same EvoLink gateway, so choose the endpoint for the answer your workflow needs. For tool calls, prefer Responses; Chat Completions supports function calling only with reasoning_effort set to none.

How are failed or refused questions handled?

Requests that fail validation return a 400 error with the failing field in param and are not charged. If the model declines one question, that answer comes back with type refusal while the other answers are returned normally.

Is the OpenAI Decisions API in beta?

Yes. OpenAI currently lists the Decisions API as public beta. That status applies to the Decisions endpoint, rather than every GPT-6 Luna request. Evaluate the question types, thresholds, and refusal handling against your own workflow before expanding usage.

Can I use the returned probability to automate routing?

Your application can use the probability to decide whether to route a request or send it for review. Set thresholds using labeled examples and the cost of false positives and false negatives; a returned probability is not a guarantee of accuracy. Keep a review path for ambiguous inputs and handle refused answers separately.