Models and effort
Switch models when you want a different balance of speed, cost and reasoning. You keep the conversation, including when you change providers.
/modelPick a model from a provider you are signed in to. The main tab changes your session model; balanced and fast choose the models used by subagents. Each model offers the effort levels it supports.
Switch during a conversation
Section titled “Switch during a conversation”Wait until the agent finishes, or interrupt it first. You can select a provider’s default model directly:
/model deepseekUse /model <provider>/<model> for a specific model, or just its model ID for the current provider. Changing models does not erase the conversation. Tatu adapts the history for the new model; a model that cannot see images needs a vision model to get their descriptions.
The catalog lists models, context windows and prices; /model refresh updates it without changing your selection.
When a provider runs out of quota
Section titled “When a provider runs out of quota”If a usage limit pauses the task, the panel offers Switch provider. It opens this same picker, limited to available models of your other providers, and continues the paused task on your choice. Closing the picker changes nothing. The choice is saved as a default under the same saveDefaults rule. See Context and cost.
Choose effort
Section titled “Choose effort”/effortChoose how hard the model thinks, or use /effort default for the model’s own default. The levels run from off (thinking turned off) and minimal up to xhigh and max, but each model offers only the ones it supports: a model that always thinks has no off, and xhigh and max appear only on the models that take them. Not every model supports effort. When you switch models, Tatu adjusts an unsupported effort to the closest level the new model accepts and tells you what changed.
By default, /model and /effort save your choices as defaults for new sessions. Turn saveDefaults off in /settings to keep changes session-only. A choice fixed by project settings is not overwritten, and an unlisted model is not saved as a default. Resuming a session restores its own model and effort without changing your defaults.