GPT-6 Luna Decisions

GPT-6 Luna Decisions is OpenAI's structured decision model, served by Decisions API through the same Jev playground and /v1/systemone endpoint developers already use for decision workloads. Where a chat model returns prose that has to be parsed, GPT-6 Luna Decisions returns typed answers with explicit probabilities, so the response is data your code can branch on immediately.

The model exposes three answer shapes. Choice returns a named option plus a full probability distribution across 2 to 255 options. Score returns a value on an ordered scale of 2 to 10 levels. Noul returns a yes/no probability. Input is multimodal: shared text, JSON objects or arrays, and images can be combined in one request, with up to 4 images per call.

Integration is deliberately boring — in the best way. You post questions and context, and read typed answers from data.result.answers. The platform caps are documented up front: 8 questions per call, 32 KiB of text plus questions, images up to 4 MiB and 16 MP each. Every rule that the team verified against OpenRouter in October 2026 — such as 255 options accepted and 256 rejected — is written down, so you know the API contract before your first call.

Because GPT-6 Luna Decisions is served through an independent aggregator, pricing stays visible: upstream cost is $0.10 per million input tokens, the platform adds 1500 credits ($0.15) per million, and minimum billing is a single credit. New accounts get 100 welcome credits, and failed calls release the credit reservation.

Typical workloads include support and agent routing, citation verification, content moderation, lead scoring, LLM guardrails, and image classification. If you have been wrapping a chat model in regex to fake structure, GPT-6 Luna Decisions replaces that glue with typed answers and probabilities — a cleaner contract for production decision systems.