GPT-6 Astra, Sol and Luna: Which OpenAI Model Does What

OpenAI completed its GPT-6 model family in September 2026. GPT-6 Astra reached the API on September 3, GPT-6 Sol and GPT-6 Luna on September 22, and an improved GPT-6.1 Sol on September 29, at $10, $2 and $0.10 per million input tokens. In ChatGPT, OpenAI's docs say the Sol and Luna models appear in Work and Codex rather than in regular chat.

Three models, three jobs

OpenAI's GPT-6 guide frames the family as a simple trade between intelligence, speed and cost. Here is how each one is positioned, with the prices from the API pricing page:

All three read text and images, answer in text, and accept up to 1,050,000 tokens of context, enough for several long books at once. The rollout came in quick steps:

The price gap is the real story

The spread between the tiers is wide. If you build automations, or pay for a tool that runs on OpenAI models, the tier you pick can change your bill a hundredfold.

Price per token is not price per job, though. OpenAI says that in several of its evaluations Astra reached stronger results with far fewer output tokens, so its cost per task came in lower than older models despite the higher rate. Independent numbers agree that Sol is the value pick: on Artificial Analysis' cost-per-task measure, GPT-6.1 Sol runs a test task for $0.72, among the cheapest of the strong models on our AI model leaderboard.

What it means if you use ChatGPT

This is where it gets confusing, because ChatGPT now has several places to work. OpenAI's ChatGPT and Codex model docs say GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna are available in Work, ChatGPT's mode for multi-step tasks with your files and apps, and in Codex, OpenAI's coding agent. They are not in regular chat.

• Plans. The GPT-6.1 Sol rollout covers Plus, Pro, Business, Enterprise and Edu. Free and Go were not included at launch, and on Enterprise and Edu an admin has to switch it on.
• The Power control. In Work and the desktop app, ChatGPT shows a Power control instead of plain model names. OpenAI's example runs from Luna High up to Astra Extra High (some paid plans omit the top setting), and an Advanced option lets you pick a specific model, effort or speed.
• Codex users. GPT-5.5 left Codex for ChatGPT sign-ins on October 14, so saved settings and scheduled tasks that named it need a new model. OpenAI suggests GPT-6.1 Sol on paid plans.

Astra works differently, so brief it differently

OpenAI's guide is candid about Astra's habits. It is more likely to stop and ask a question when the answer could change the result, it leans toward lists, tables and heavy formatting, and it pays closer attention to instruction files such as skills. None of that is bad, but it surprises people who expect the old behavior. A short standing instruction fixes most of it:

Do the work I ask for without stopping to confirm small details; make sensible assumptions and list them at the end.
Ask me a question only if the answer would change the result in a big way.
Write in short plain paragraphs. Use a list or table only when I am comparing things.
Before you send, check every number against the source I gave you.

If you are choosing between OpenAI's and Anthropic's coding agents, our Claude Code vs Codex comparison walks through setup, cost and which suits which kind of user.

What to watch next

• Old models shutting down. OpenAI retires a batch of legacy API snapshots and the original gpt-image-1 image model on October 23, and Agent Builder on November 30. Tools built on them will have to switch.
• Free users. OpenAI has not said when Free and Go plans get GPT-6.1 Sol.
• The human vote. Astra leads several hard benchmarks but sits only 30th on LMArena's text board, where people pick answers blind. Watch whether its chat style improves in later snapshots.