Choosing a model

Each assistant runs on a model. You choose it with a single control that offers two ways to decide.

Tier or specific model

• A tier: pick a quality/cost level (for example, value, performance, or frontier) and let the environment map it to a real deployment. This keeps assistants portable: the same assistant works across environments that map the tier to their own models.
• A specific model: pin a named model when you need exactly that one.

Choose a tier for most assistants; pin a specific model only when there’s a reason to.

Note

Which tiers and models are available, and how tiers map to real deployments, is set by your administrators under Models. If a tier isn’t mapped yet, an administrator needs to map it before the assistant can run.

Other model settings

Alongside the model you can adjust settings such as temperature (how creative versus precise the replies are). Sensible defaults are provided, change them only when you have a clear reason.