---
title: 'GPT Luna'
description: "OpenAI's named GPT tier for efficient, focused workloads, including high-volume tasks and lower-cost steps within agent systems."
canonical_url: 'https://darkfactory.dev/glossary/luna'
markdown_url: 'https://darkfactory.dev/glossary/luna.md'
collection: glossary
date_published: '2026-10-07T00:00:00-04:00'
date_modified: '2026-10-07T00:00:00-04:00'
---

# GPT Luna


## Definition

Luna is OpenAI's named GPT tier for efficient, focused workloads. It can supply lower-cost steps in an agent system, such as a bounded classification or a routine coding task. The name spans distinct numbered releases and does not guarantee that every task is suitable for the same model.

## Origin

OpenAI introduced GPT-5.6 Luna alongside Sol and Terra in the June 26, 2026 limited preview. GPT-6 Luna was announced on September 22. The current model documentation identifies `gpt-6-luna` as a model for focused, high-volume tasks.

## Interfaces and capability

GPT-6 Luna supports text input and output, image input, and configurable reasoning effort. Its documented tool-calling rules differ between the Responses and Chat Completions APIs. A product exposing a Luna model may add retrieval, computer use, or other tools; the product's permissions still govern those actions.

OpenAI's October 6 Decisions API beta uses GPT-6 Luna for predicate, choice, and score outputs. That deployment illustrates one use of Luna. It does not make all Luna output a decision primitive or make Luna and decision model interchangeable names.

## Operational significance

A cheaper call can become an expensive workflow if it causes more retries or review. Test the intended workload, then define escalation conditions for a different model. A model cascade needs an acceptance rule that can recognize insufficient output; low confidence and a failed verification check provide different evidence.

OpenAI's September release comparisons vary reasoning effort and measure particular task sets. Reported savings and capability need to be tested on the deployment's own workload.

## Distinguish it from nearby terms

- Sol and Astra are separate GPT tiers, with different intended tradeoffs.
- A small language model is a scale category. The Luna name does not disclose a parameter count or prove membership in that category.
- A decision model returns bounded judgments under a defined output contract. Luna also supports other uses.
- Model routing is the application's selection process; Luna is one possible destination.

## Check your understanding

A workflow routes all requests to Luna to reduce its bill. What measurements would show whether retries, escalation, and human review have increased the cost per successfully completed task?

## Related terms

- [GPT](https://darkfactory.dev/glossary/gpt)
- [GPT Sol](https://darkfactory.dev/glossary/sol)
- [GPT Astra](https://darkfactory.dev/glossary/astra)
- [GPT Terra](https://darkfactory.dev/glossary/terra)
- [Model routing](https://darkfactory.dev/glossary/model-routing)
- [Decision model](https://darkfactory.dev/glossary/decision-model)
- [Reasoning model](https://darkfactory.dev/glossary/reasoning-model)

## Related factory areas

- [Model selection, routing & budgets](https://darkfactory.dev/factory/model-routing-budgets)
- [Economics, capacity & factory FinOps](https://darkfactory.dev/factory/economics-finops)
- [Verification, evaluation & quality truth](https://darkfactory.dev/factory/verification)

## Evidence and further reading

- [OpenAI: Previewing GPT-5.6 Sol](https://openai.com/index/previewing-gpt-5-6-sol/)
- [OpenAI: Introducing GPT-6 Sol and Luna](https://openai.com/index/introducing-gpt-6-sol-and-luna/)
- [OpenAI API: GPT-6 Luna model](https://developers.openai.com/api/docs/models/gpt-6-luna)
- [OpenAI API: Decisions](https://developers.openai.com/api/docs/guides/decisions)
- [OpenAI API: Model selection](https://developers.openai.com/api/docs/guides/model-selection)