Chat Model Selection
How to choose the model used for text chat in the Web App.
Overview
In the Web App, the model in the header is the Chat model. It handles text messages, attachments, tool use, image generation, web search, and thinking mode in chat.
The Chat model is separate from Settings → Voice. Voice Agent has its own model and voice choices for Standard mode and Realtime mode. See Model and Voice Settings.
Switching Models
Click the model name in the header to open the chat model selector. The list shows every active model: plans do not restrict models, and more capable models simply use more credits.
On desktop, hover over a model to see:
- Credit usage
- Model description
- Supported input types
- Supported output types
See AI Models for the full catalog and how the default is chosen.
When to Change the Chat Model
| Need | How to choose |
|---|---|
| Fast answers and daily tasks | Use an efficient model or keep the default |
| Long analysis, reasoning, or coding | Choose a stronger reasoning model |
| Files or multimodal content | Choose a model with the right input/output types |
| Cost control | Check the credit usage label in the dropdown and see Credits & Usage |
Thinking Mode
Some Chat models support thinking levels. When the current model supports it, a Think selector appears in the input bar.
| Level | Best for |
|---|---|
| Instant | Fast responses without extended reasoning |
| Thinking | Multi-step questions and analysis |
| Thinking Extended | Difficult math, logic, coding, or long analysis |
If the selector does not appear, the current Chat model does not support this mode.
The available levels come from the selected model. Models can expose fewer levels or no thinking selector at all.
Default Model and Remembered Selection
The model marked Default in the list is used when you don't pick one: the LLM you set in Settings → Voice, else the assistant's model, else the platform default.
- Picking another model is remembered on this device for future conversations.
- Picking the Default model again stops pinning, so the header follows the default and updates when the assistant owner changes it.