Getting started
AI models
Which AI model writes your app: what Auto picks on each plan, how to pick one yourself, what each plan includes, and what happens when a model is unavailable.
Seth writes your app's code with an AI model from Anthropic (Claude), OpenAI (GPT) or Google (Gemini). You can leave the choice to Auto, or pick a model yourself from the menu under the box where you write, next to +. The button shows the model in use.
Auto
Auto is the first row of the menu: the best model for each request, picked for you within your plan.
| Plan | What Auto uses |
|---|---|
| Free | GPT-5.4 mini, the model Free builds with |
| Pro | Claude Sonnet 5.5, for every request |
| Business | Claude Fable 5.1 for a game, Claude Opus 4.8 for a complex app, Claude Sonnet 5.5 for the rest |
On Pro, Auto stays on Sonnet so the credits go as far as the plan promises. The stronger models are there to pick by name.
With Power Seth or Max Seth chosen, Auto moves up with the mode: both Power Seth and Max Seth build on Claude Opus 5.5. To build on Claude Fable 5.1, pick it by name: a model you pick by name stays the one that builds, and the line under the box shows its price first. See Build modes.
That line also says which model the mode runs on, for example "Runs on Opus 5.5." when it is Auto, or "Runs on Fable 5.1, as picked." when you chose it. Beside it, Pick another model (or Change model, once you have picked one) opens this menu, so you can change the model without leaving the line.
Pick a model yourself
The menu lists the models by who makes them, the newest and strongest first. Each row says what it means for you:
- apps left with your credits, on the dashboard: how many new apps your balance still buys on that model. In the editor the same row says messages left: how many changes it still buys. A stronger model costs more credits, so its number is smaller.
- what Auto uses here: the model Auto already runs on your plan.
- A crown with Pro or Business: the plan that opens that model.
Your pick is remembered on this browser for every app, until you change it. You can change it before any request. If your plan changes and your pick is no longer included, you are back on Auto.
| Free | Pro | Business | |
|---|---|---|---|
| Auto | Yes | Yes | Yes |
| Pick a model | Yes | Yes | |
| Claude Sonnet and Opus, GPT-5.x, Gemini | Yes | Yes | |
| Claude Fable 5.1 and GPT-6 Astra, the strongest | Yes |
What each model costs
Credits per request, as they stand today. The same figures decide how many apps and messages the menu says your credits still buy. The price on the button is always the one to trust.
Claude, from Anthropic
| Model | A new app | A change | A Chat reply | Plan |
|---|---|---|---|---|
| Fable 5.1 | 245 | 75 | 13 | Business |
| Opus 5.5 | 125 | 40 | 7 | Pro |
| Opus 4.8, 4.7 and 4.6 | 125 | 40 | 7 | Pro |
| Sonnet 5.5 | 50 | 15 | 3 | Pro (Auto) |
| Sonnet 4.6 | 75 | 25 | 4 | Pro |
Fable 5.1 is the strongest model Setherion has. On Business, Auto picks it for a game. A change on it costs about twice a change on Opus. Hover a model in the menu to read its short note.
GPT, from OpenAI
| Model | A new app | A change | A Chat reply | Plan |
|---|---|---|---|---|
| GPT-6 Astra | 245 | 75 | 13 | Business |
| 5.6 Sol | 100 | 30 | 5 | Pro |
| 5.6 Terra | 60 | 20 | 3 | Pro |
| GPT-5.5 | 140 | 40 | 7 | Pro |
| GPT-5.4 | 70 | 20 | 4 | Pro |
| 5.4 mini | 25 | 10 | 1 | Every plan (Auto on Free) |
GPT-6 Astra is OpenAI's top model. Auto never picks it: it runs only when you choose it. When something keeps going wrong on one family of models, trying another family sometimes helps.
Gemini, from Google
| Model | A new app | A change | A Chat reply | Plan |
|---|---|---|---|---|
| 3.1 Pro | 60 | 20 | 3 | Pro |
| 3.5 Flash | 45 | 15 | 2 | Pro |
A new app, a change and a Chat reply are priced on the model that does the work. Power Seth and Max Seth add to a new app and a change: see Build modes.
If Setherion opens a new app for a first check and finds it crashing, the repair is done by Claude Fable 5.1 whatever plan or model you chose, and it costs you nothing.
Which one to choose
- Leave it on Auto for most work. It is what the prices on the pricing page are worked out on.
- Pick a stronger model for a big app with many screens, or a game, when the first try on Auto missed something.
- Pick a fast one, like Gemini 3.5 Flash or GPT-5.4 mini, for small changes: a word, a colour, a section moved. It does them for half the credits or less.
Different models write differently. If one keeps getting something wrong, the same request on another often gets it right.
When a model is unavailable
When a model does not answer, the work is not lost: it carries on with a model at least as strong from another maker, and the line under the work says so while it happens, for example "Anthropic is unavailable, continuing on OpenAI". If the other model wrote the code, the finished result says it too: "Claude Sonnet 4.6 was unavailable, so GPT-5.6 Sol built this app." You are charged the price of the model you chose, never more.
The models change as the makers release new ones, and the menu always shows what your plan has today.
Was this page helpful?
