GPT-6 Luna logo

GPT-6 Luna

LLM

GPT-6 Luna is the fast, high-volume tier of OpenAI's GPT-6 family, released 22 September 2026 alongside Sol and priced 50% below its GPT-5.6 predecessor. It is built for repeatable work at scale: classification, extraction, labelling and short replies, with reasoning that can be switched off entirely for the lowest latency. Free and Go users get it in the desktop app.

What GPT-6 Luna is best for

High-volume classification, extraction, tagging, routing and short-form generation; first tier in a two-tier setup where hard cases hand off to Sol or Astra; free-tier users who want GPT-6 behaviour in the desktop app.

How to use GPT-6 Luna

Set model to gpt-6-luna and reasoning_effort to none for the fastest, cheapest turns, or low when a field must be inferred. Use strict tools or json_schema output so results are machine-safe. Drop temperature and top_p when reasoning is on. Batch independent records and run them in parallel.

How it behaves

Optimised for speed and volume. Supports reasoning_effort none, where all judgment must live in the instructions and the output should be a strict function call or JSON schema. Function calling on Chat Completions works only at none; use the Responses API for reasoning with tools. Follows instructions tightly and inherits the GPT-6 bias toward action.

Tips

Write the label boundaries into the system prompt and force a strict function call so the code can trust the output. Ask for a short evidence quote alongside every label as an audit trail. Keep effort at none or low for copy-and-extract work and escalate the uncertain cases to Sol.

Pricing

Included for Free and Go users in the ChatGPT desktop app, and in paid plans. API prices are 50% below GPT-5.6 promotional pricing; see the OpenAI pricing page for current per-token rates.

Prompts for GPT-6 Luna

All 7 GPT-6 Luna prompts

Bias Toward Action Operator Preamble

Generate an action-oriented assistant prompt focused on task completion and minimal clarification requests, outputting a structured report of actions and assumptions.

31

Concrete Draft Before Clarifying Questions

Generate a complete working draft before asking clarifying questions, ensuring assumptions are clearly marked.

29

User Instruction Precedence Over Loaded Skills

Defines instruction precedence for a workspace assistant in interactions within a specified workspace.

22

High-Volume Classification Run With No Reasoning

Generate consistent, high-volume labels for records without reasoning, using structured function calls.

21

Plain Paragraph Output Policy

Generate clear and concise response templates through structured writing rules, ensuring clarity in discussions.

17

Batch Extraction Schema for Repeatable Work

Generate structured JSON output for data extraction from specified document types, ensuring accurate field replication.

12

Alternatives to GPT-6 Luna