Skip to main content
Custral works with multiple AI models from OpenAI and Anthropic. The same chat or agent can run on any of them. Pick the model that fits the job, and switch at any time without losing your setup.

Catalog

Context window is how much text the model can consider at once. Tools indicates whether the model can take actions (read records, send email, search the web). Vision indicates whether it can read images. The default model for a new workspace is GPT-4o.

Tiers

Each model belongs to a tier that signals its trade-off between speed and depth:

Fast

Quick, low-cost responses. Best for high-volume property auto-fill and simple chat. (GPT-4o mini, Claude Haiku 4.5)

Balanced

The safe default for most chat and agent work. (GPT-4o, Claude Sonnet 4.6)

Reasoning

Deeper reasoning for complex chat and schema design. (Claude Opus 4.7, o1-mini)

Which model runs

When you send a message, Custral picks the model to use in this order:
1

The model you pick in chat

A model chosen directly in the chat takes precedence over everything else.
2

The conversation's model

Once a conversation starts, its model sticks for the life of that conversation.
3

The agent's model

A chat agent can declare a preferred model. Set it to Auto to defer to the workspace default.
4

The workspace default

Set on the workspace AI settings page, this applies when nothing more specific is chosen.
5

Fallback

If no preference applies anywhere, Custral uses GPT-4o.
Some background tasks (like generating a short description) always use a fast, low-cost model regardless of the choice above.