Change the model without rebuilding the workspace.

Use a representative task with the same source material, expected output, and acceptance criteria. Compare factual accuracy, time, and the amount of revision you need rather than relying only on a benchmark ranking.

Auto is the name used in the model directory; the current user guide calls the automatic routing entry Lite mode. An explicit model choice fixes the requested target, while automatic routing delegates that choice. Confirm the names and entries shown in your client.

Different models use different execution paths. The directory separates chat selection from image and video models accessed through built-in Combos; absence from the chat list alone does not mean the model is unavailable everywhere.

No. A configuration record, a public pricing listing, and a selectable client entry are different kinds of evidence. The directory records them separately, because the online menu and account eligibility can change.

Not necessarily. What matters is how well the model uses the relevant evidence and follows the assignment. Keep sources organized, test retrieval of critical details, and check whether larger inputs improve the accepted result.

Trace which files or extracted content are sent to the selected provider and what connected tools receive. Local files and a desktop interface do not, by themselves, establish that inference stays on the device.