Choose and manage models

Picking a model

The model picker groups entries by provider connection. Choose the model before sending a message. You can switch during a conversation; each response retains the name of the provider and model that produced it.

A switch sends the relevant conversation context to the newly selected provider. If you are working with sensitive information, consider that transfer before switching. Existing images or PDFs must be compatible with the new model.

Capability labels

Vision means image input is supported. Document support refers to native PDF input in an implemented adapter. Tool support allows the model to request the workspace's enabled tools. Labels are based on documented model information or explicit manual configuration, and are not promises for every model offered by a provider.

Generation settings

Temperature controls sampling variety for models that support it. Maximum output tokens limits response length. Top-p is another sampling control. Some reasoning models ignore or reject sampling settings, so the adapter omits unsupported settings for known model families. A provider remains the authority on its accepted parameters.

Usage and costs

Usage is shown when the provider reports token counts. Missing usage is shown as unavailable rather than treated as zero. Costs are estimates only when reliable model pricing is returned. Your provider dashboard is the authoritative billing record; caching, reasoning tokens, discounts and plan-specific pricing can change charges.

Comparing responses

Open comparison, select two models, and enter a prompt. Review each response on its merits. The requests are independent and can each incur charges. A faster or longer answer is not necessarily a better answer.

Refreshing a list

Fetch models after enabling new account permissions or when a saved ID stops working. Manual entries need your own review. Removing a connection keeps historical response labels but prevents new requests through that connection.