Choosing the Right Model

By Task

You want to	Use	More
Generate or discuss text	A chat model	Model Families
Work a problem step by step	A reasoning-capable model	Reasoning
Ask about an image	A vision-capable model, with `--file`	Image Input
Ask about a document	Any model, with `--file`	PDF Input
Pull in current information	Any model, with web search on	Web Search
Compare several answers	Fusion	Fusion

Smaller models are faster and cheaper, and usually enough. Reasoning costs extra tokens and time, so save it for problems that genuinely need the steps.

By Privacy Level

The level rides with the model you pick, so this is the same act as choosing the model. Take the lightest level that fits the call.

Level	Take it when	What it gives you
Anonymous	Routine work	Frontier models, with your identity detached
Private	You want no data retained	Zero data retention enforced upstream
Confidential	You want attested privacy	Sealed execution with an attestation you can check

A call that cannot be served at the level you asked for fails rather than quietly dropping to a weaker one. Full ladder: Privacy Levels.

Then Set It

models browses the catalog, model sets a default, --model overrides one call. Syntax is on Choosing a Model. None of it changes the model your harness runs.

Model List

The current catalog, grouped by privacy level.

Choosing a Model

The command syntax for browsing and setting your model.

Pricing

How cost is structured and where to read per-model rates.

​By Task

​By Privacy Level

​Then Set It

​Related Resources