Which Agent settings and model should I use?

Write instructions around the task, sources, expected output, and limits. Start with a generally capable model available in your workspace, then compare results on a small test set. Choose a more capable or specific model only when the task requires it and workspace policy allows it.

Model options and reasoning controls vary. A per-conversation model choice may affect only that conversation; changing the Agent’s saved model affects future use. Check the setting description before saving.