There is a cost of your data / prompt + output being collected by the upstream providers for training if you use them:

> DeepSeek V4 Pro: smartest. Its API collects data for training.

> DeepSeek V4 Flash: most efficient. Its API also collects data for training

> MiniMax M3: smartest unlimited model, multimodal. Its API collects data for training.

There needs to be an option to use local models instead to guarantee that no data would ever leave your machine or stored by someone else's machine for training purposes.