GPT-6 Astra and GPT-6 Luna: OpenAI models over the API, no VPNAI models by name: what each one does and where to switch it on

Updated

GPT-6 Astra and GPT-6 Luna: OpenAI models over the API

Both OpenAI models are connected here, billed from your shared balance with no separate OpenAI account. Two caveats come first, because without them the page would be misleading: these are models over an API, not the ChatGPT app, and Twelver’s chat does not answer on them. What you actually get is below.

If you were looking for the ChatGPT app

It is not here and will not be. ChatGPT is OpenAI’s product with its own interface, memory, image generation and everything else; what we provide is the model that works inside it, through an OpenAI-compatible endpoint. They are different things and we are not going to substitute one for the other.

In practice: you call gpt or gpt-max from your own code, or from any client that speaks the OpenAI API — Cursor, Cline, opencode and the rest. Those clients are covered in the coding API book.

By default the chat does not answer on GPT

A model is picked automatically for every answer, and right now that automation runs the conversation on GLM, not GPT. GPT-6 Luna can be pinned to an agent: in its settings, under “Text model”, choose “ChatGPT” and save — choosing the model is part of the Pro plan.

The flagship Astra is not offered in the chat, and that is deliberate: nothing routes to it, and it answers only whoever names it over the API.

Two models, a hundredfold apart in price

Both get the same very large context — roughly a million tokens — and the same ceiling on a single answer. They differ in quality and in price, and the price difference is severe.

What to pass in modelWhich model answersIn / out, $ per 1M
gptGPT-6 Luna0.163 / 0.813
gpt-maxGPT-6 Astra16.25 / 81.25
Prices in dollars per million tokens, reconciled with /api/pricing on 22 October 2026. Billed in Twelver tokens — no separate OpenAI account required.

The gap is exactly a hundredfold, on input and on output alike. That is the widest distance between two tiers of one vendor anywhere in this book, and for jobs where the model writes a lot — code generation, long answers, an agent — it decides the bill outright.

Start on Luna, not the flagship

Reaching for gpt-max because “it is the flagship” is expensive and usually changes nothing: on ordinary work — rewriting text, answering from a document, parsing data — you will not see a difference in the result, and you will pay a hundred times more for every answer.

The sensible order: do the job on gpt, see exactly where the result falls short, and only then try gpt-max — on that specific point.

~1MTokens of contextOn both models: a large document or a whole repository
128KMaximum answerAlso on both — the ceiling of a single model response
$0.813 / $81.25Output per 1M tokensLuna against Astra — exactly a hundredfold, on input too

Reasoning: on Luna it can be switched off entirely

Both models can “think” before answering, and the depth is adjustable. But the smaller one has a mode the flagship does not: reasoning turns off to zero — and that is verified, the answer really carries no reasoning tokens at all.

ModelReasoning depthWhat it means on the bill
gptFrom fully off to maximumOn simple work reasoning can be removed entirely and costs nothing.
gpt-maxFrom low to the highest setting; cannot be switched offEven the simplest request pays for some amount of reasoning.
Reasoning tokens are billed as ordinary output, so on high-volume work switchable reasoning saves more than the difference in rate.

Without leaving the page

Get a key and call GPT-6 Luna

The button creates a key and drops it into the example. Then change model to gpt and the request goes to GPT-6 Luna; gpt-max calls GPT-6 Astra.

Press it and get your base_url and API key.

No sign-up, no card.

Free. Your key appears right here.

Which GPTs are not here

OpenAI’s line-up is wider than what is connected, and pretending otherwise would be dishonest.

  • The 5.6 generation, GPT-5.6 Luna included. Thegpt value now answers on GPT-6 Luna — newer and half the price — and 5.6 models cannot be named in model.
  • The ChatGPT app and everything in it: memory, voice mode, image generation inside the conversation. We do have images, video, music and speech — but other models make them, each its own.
  • Claude and Gemini as text models. Not available at all — neither in chat nor over the API. What to use instead is covered in the comparisons book.

Frequently asked questions

Can I use GPT without an OpenAI account?
Yes, for the models. Requests go to our endpoint rather than straight to OpenAI’s servers, so no OpenAI account is needed and billing comes from your shared balance. But this is about the models over an API, not the ChatGPT app, which we do not have.
How does GPT-6 Astra differ from GPT-6 Luna?
Context and maximum answer length are the same — about a million tokens and 128K respectively. They differ in quality and price: Astra costs exactly a hundred times more on both input and output. For most work there will be no difference in the result, so starting on Luna is sensible.
How do I reach GPT-6 Astra?
Pass gpt-max in the model field when calling our OpenAI-compatible API. Nothing routes to this model automatically — it answers only whoever names it explicitly. For GPT-6 Luna it is the same with the value gpt.
Why does the chat not answer on GPT?
Because by default a model is picked for each answer automatically, and the conversation currently runs on GLM. That is a deliberate setting, not a fault. GPT-6 Luna can be pinned to an agent: its settings have “ChatGPT” under “Text model” (Pro plan). GPT-6 Astra is API-only, as gpt-max.
What does GPT cost over the API?
The prices are in the table above, in dollars per million tokens, reconciled with our own price endpoint. No separate OpenAI account is needed — it is deducted from your Twelver balance.
Can I switch reasoning off and not pay for it?
On GPT-6 Luna, yes, completely: in that mode the answer carries no reasoning tokens at all. On GPT-6 Astra it cannot be switched off entirely — a minimum level is always spent. Reasoning tokens are billed as ordinary output.
Do Cursor, Cline and other editors work with this API?
Yes — it is an ordinary OpenAI-compatible endpoint, so any client that lets you change the base URL and key will work. Setting up specific editors is covered in the coding API book.

How to connect a model to a code editor is in the coding API book. Other models are in the book’s table of contents; what to use in place of the ones we do not run is in the comparisons book.

Try it yourself

Everything in this guide runs inside Twelver

One chat for text, images, video, music and voice — no separate services or subscriptions.

Open Twelver chat
Related pageTwelver AI Chat
Rate your experience