Models and settings
The App Builder Assistant can run on different AI models. You choose which one, and you can change it at any time without losing your conversation. This page covers the model picker, the settings around it, and how usage is billed.
Choosing a model
Section titled “Choosing a model”A model picker sits next to the message box. It lists the available models grouped by provider, marks the current default, and shows a relative cost multiplier next to each one, so you can trade capability against cost with your eyes open.
Switching model mid-conversation is instant: the thread continues, now answered by the model you picked.
Available models
Section titled “Available models”Several models are available today, from more than one provider. Lighter models are well suited to quick questions and small edits, while stronger ones handle deeper reasoning and more involved building work.
| Provider | Model | Best for |
|---|---|---|
| Anthropic | Claude Haiku 4.5 | Fast and economical: quick questions and small edits. |
| Anthropic | Claude Sonnet 4.6 | Balanced, strong all-round building; a solid everyday default. |
| OpenAI | GPT-5.3 Codex | Writing and refactoring code. |
| DeepSeek | DeepSeek V4 Pro | Deep reasoning, coding and agentic work, with a very large context at low cost. |
| DeepSeek | DeepSeek V4 Flash | The most economical option, for lighter tasks. |
| Kimi (Moonshot) | Kimi K2.6 | Long-context work; also understands images. |
The exact line-up evolves as new models ship. What you see in the picker is always the current, up-to-date set.
Settings and defaults
Section titled “Settings and defaults”You decide which models are available and which one is the default, at two levels:
- Per project, from the project settings, so a given app can standardize on a chosen model.
- Per account, from your user settings, so your preference applies across your projects.
Enable the models you want to use, disable the ones you do not, and set the default that new conversations start with.
How usage is billed
Section titled “How usage is billed”ALIGN has no subscription. You add credit to your account and it is drawn down as you use the platform, so you only pay for what you actually run.
AI usage is metered by tokens, the units of text a model reads and writes. That is why each model shows a cost multiplier: a heavier model spends your credit faster than a lighter one for the same task. Picking the right model for the job is the simplest way to keep usage efficient.