Models
GPT-6 Luna
GPT-6 Luna is OpenAI's quickest and cheapest GPT-6 model. It is the default model in GBT Chat and in ChatGPT's free plan, reads photos, and answers most everyday questions in a second or two.
Luna in short
GPT-6 Luna is the smallest and quickest member of OpenAI's GPT-6 generation. OpenAI released it on 22 September 2026 together with GPT-6 Sol, and on its own price list describes it as the model for "fast and efficient everyday work at scale". In practice that means short waits, low cost and solid answers to the questions most people actually ask.
Luna matters for two reasons. It is the model that answers in ChatGPT's free and Go plans, where it replaced the previous default on 8 October 2026. And it is the default model in GBT Chat, so it is very likely the one that answers when you type on our home page.
- Reads: text and images; writes text.
- Context window: 1,050,000 tokens at OpenAI, with up to 128,000 tokens of output.
- Knowledge cut-off: 18 May 2026, according to OpenAI's model documentation.
- API price: $0.10 in and $0.50 out for each million tokens (OpenAI list price, Nov 2026).
Luna's place among the GPT-6 tiers
Since mid-2026 OpenAI names its models with a generation number and a tier. In the GPT-6 generation the tiers are, from top to bottom: Astra (the flagship, 3 September), Sol (22 September, upgraded to GPT-6.1 Sol on 29 September), Terra (GPT-6.1 Terra, 26 October) and Luna. Each step down is faster and cheaper; each step up handles harder, longer work.
The price gap shows the positioning. Per token, Luna costs a tenth of GPT-6.1 Terra and a twentieth of GPT-6 Sol. If you remember older names, Luna fills roughly the role that "nano" models used to: the one you choose when speed and volume count more than depth. That does not make it weak. It is a current-generation model with a large context window and image input, and for everyday tasks the difference from the bigger tiers is often hard to notice.
The details are on OpenAI's GPT-6 Sol and Luna announcement.
What Luna does well
- Everyday writing. Messages, e-mails, captions, short posts, polite refusals, a birthday note. It follows tone instructions well: "warmer", "shorter", "less formal".
- Quick explanations. What a word means, how something works, the difference between two ideas, a summary of a page you paste in.
- Light study help. Flashcards, quiz questions, step-by-step explanations of school-level topics. Our quiz generator is an easy way to try this.
- Reading photos. A sign or menu in another language, a handwritten note, a chart, the label on a product. Attach the picture and ask.
- Translation and rewording between widely used languages, including keeping the meaning while changing the register.
- Simple code questions: what a short snippet does, a regular expression, a spreadsheet formula.
Where it falls short
Luna is tuned for speed, so it spends little effort on hidden reasoning. You will notice that on:
- Multi-step problems: tricky maths, logic puzzles, plans with many constraints.
- Long, careful analysis: comparing several documents, reviewing a contract clause by clause, debugging a larger program.
- Subtle writing where every sentence has to be right, such as a sensitive letter.
Like every language model it can also state wrong facts confidently, and it knows nothing after mid-May 2026. For those harder jobs in our chat, press Think to switch to GPT-6.1 Terra; the Luna vs Terra comparison shows where the switch pays off.
How GBT Chat runs Luna
On our site, Luna answers without any extra thinking step, which keeps replies fast. It reads one attached image per message. In long conversations the chat sends the model the most recent part, so very early details may drop out; starting a fresh chat for a new topic helps. Answers are free within a few messages a day for every visitor, and your history stays in your own browser.
A practical note on privacy: your message passes through our server to the model and back, and we do not keep the text of your chats. Even so, treat Luna like any online service and leave out passwords, card numbers and private documents. If you want to keep a conversation, export it from the chat menu or create a share link; clearing your browser data removes the local history for good.
Nothing to set up: Chat with GPT-6 Luna now, or start with one of these:
Getting better answers from Luna
- Give it the material. Paste the text, numbers or message you are asking about rather than describing them.
- Ask for a format. "Three bullet points", "a table with two columns", "under 80 words".
- Break big jobs into steps. Ask for an outline first, then each part.
- Ask it to check itself. "List anything in your answer you are not sure about" often brings weak points to the surface.
Our first chat guide walks through these habits with examples.
FAQ
Is GPT-6 Luna the same model as in free ChatGPT?
Yes, the same OpenAI model. Since 8 October 2026 ChatGPT's Free and Go plans use GPT-6 Luna for everyday chat. The instructions and settings around it differ between services, so answers can vary in style and length.
Is GPT-6 Luna free to use?
In GBT Chat, yes, within a daily allowance and without an account. In ChatGPT it is the model of the free plan. Developers pay per token through the API.
Can GPT-6 Luna read images?
Yes. It accepts images as input and answers in text. In our chat, tap the plus button to attach one photo per message.
Can Luna create pictures or search the web?
No. The model writes text only. ChatGPT adds search and image creation as separate tools; our chat has neither.
How recent is Luna's knowledge?
OpenAI gives 18 May 2026 as its knowledge cut-off. It does not know about later events unless you tell it.
Keep reading
Your question, answered now
No account, no card, nothing to download. Type what you want to know and the first words of the answer appear in a second or two.