GPT-6 Sol and Luna Cost-Efficient Models
Released by OpenAI on 22 September 2026 and trained with methods similar to GPT-6 Astra to bring frontier intelligence to faster, cheaper models, with API prices cut 50% versus GPT-5.6 promotional pricing
Tool Overview
Features, steps and FAQ below
Features
- ✓ Price: OpenAI says GPT-6 Sol drops from $4/$20 to $2/$10 per million input/output tokens and Luna from $0.20/$1.20 to $0.10/$0.50, both 50% cheaper than GPT-5.6 promotional pricing
- ✓ Factuality: OpenAI says Sol makes about half as many mistakes as its predecessor on its internal factuality evaluation, approaching Astra-level reliability at far lower cost, while Luna improves substantially at higher effort
- ✓ Coding: OpenAI says Sol improves substantially over GPT-5.6 Sol on FrontierCode and scores 68.8% at max effort on DeepSWE v1.1, within 1.1 points of Fable 5's best at about 80% lower cost per task
- ✓ Computer use: OpenAI says on OSWorld 2.0 Sol matches Claude Opus 5 at medium effort at roughly 80% lower cost per task, while Luna exceeds GPT-5.6 Sol at about one tenth its cost
- ✓ Collaboration style and caching: Astra's clearer communication with less jargon and fewer low-value details, plus improved caching for agents and long conversations
How to Use
- Pick GPT-6 Sol or GPT-6 Luna in ChatGPT Work and Codex, available to most paid accounts
- On the ChatGPT API set the model name to the corresponding GPT-6 Sol or Luna model
- Luna is also available in the desktop app and for Free and Go users, so you can try it cheaply first
- Tier by difficulty: use Astra for your most demanding work and Sol or Luna for everyday tasks at scale to cut cost
FAQ
What are GPT-6 Sol and Luna?
Two models OpenAI added to the GPT-6 universe on 22 September 2026, trained with methods similar to GPT-6 Astra to bring its advances in frontier work to faster, cheaper models.
How much did prices drop?
OpenAI says 50% versus GPT-5.6 promotional pricing: Sol is $2 input and $10 output, Luna is $0.10 input and $0.50 output per million tokens.
How do they differ from Astra?
OpenAI says Astra remains its best model across the board for the most demanding and important projects, while Sol and Luna focus on cost efficiency so frontier capability becomes practical for everyday tasks at scale.
Are they more accurate?
OpenAI says Sol makes about half as many mistakes as its predecessor on its internal factuality evaluation and gets close to Astra-level reliability, while Luna improves substantially at higher effort levels.
Where are they available?
From launch they are in ChatGPT Work and Codex for most paid accounts and in the ChatGPT API, with Luna also coming to the desktop app and Free and Go users.
How good are they at coding?
OpenAI says Sol improves substantially over GPT-5.6 Sol on FrontierCode and scores 68.8% at max effort on DeepSWE v1.1, within 1.1 points of Fable 5's best at about 80% lower cost per task.