Choose the model and reasoning effort

Every chat runs on a model, and the reasoning effort sets how much that model thinks before it answers. This page covers the model picker in the message box and the /model command: what the list shows, how to change the model and the effort, and what its messages mean.

Open the model list

• The model control sits in the message box's control row. It shows the model's mark, its name and its effort level; when the row runs out of space it shortens to the mark alone, and the full choice stays in its tooltip and in the open list.
• The /model command opens the same list for the chat.
• The two always agree: a change made in one is what the other shows next time it opens.

Choose a model

1. Open the model list: click the model control in the message box, or type /model.
2. Pick a model. If the list holds more than six models, a Search models box appears, and typing narrows the list.
3. The list closes. From the next request, Oda uses the model you picked.

• The change applies from the next request. A request that is already running keeps the model and effort it started with.
• Picking the model that is already set just closes the list; no request is sent.
• Each row shows the model's description, and a New badge on a recent addition. Rows are grouped by the company that makes each model: DataMind's own models first, then the other companies in the order the list gives them.
• Some rows also show a size such as 1M; DataMind's own models never show one, by design.
• A model that is hidden, or one that leaves the list, stops appearing here. Model selection and defaults covers hidden and unavailable models, and Models in the chat picker is where hiding is set.

Choose a reasoning effort

1. Open the model list and highlight the model you want; the Effort control at the foot of the list follows the highlighted model.
2. Choose a level. It applies at once, and the list stays open.

• The levels are Off, Minimal, Low, Medium, High, Extra high and Max, as far as the model supports them; a model that offers no levels shows no Effort control. Default appears where a model carries no default of its own.
• Changing the level for a model the chat is not using switches the chat to that model.
• A request that is already running keeps the effort it started with.

From the keyboard

• The list opens on the chat's model.
• Up/Down, Home/End and PageUp/PageDown move through the rows.
• Type any letter or number to search the list.
• Enter picks the highlighted model and closes the list. Tab moves between the rows and the Effort control.
• Left/Right change the effort level.
• Escape, or Shift+Tab from the rows, closes the list back to the model control. A click outside closes it as well.

When the list will not load

• While Oda reads the list, the control reads Loading models…. When it reads the list again, the last list stays visible and the menu notes Refreshing model list….
• If the whole list cannot be read, the menu shows Catalog failed to load: {message} with Retry.
• If one company's models cannot be read, its group shows {name} failed to load: {message} with Retry; the groups that loaded stay usable.
• A row Oda cannot resolve is dimmed, and you cannot pick it.
• An empty list reads No models available., and a search with no match reads No model matches your search..
• While a switch is in flight, its row shows Switching model…, and effort presses wait for it. A switch that fails shows Model operation failed: {message} and leaves the list open, so you can pick again; your chat keeps the model it had.
• If another running copy of Oda is using the chat, the change is refused and Oda asks you to quit the other running instances; The message box explains that message.

Questions, answered

Is /model different from the model picker?

No. Both open the same list for the chat, so a change made in one is what the other shows next; it applies from the next request.

Why are the models grouped by company?

Rows are grouped by the company that makes each model. DataMind's own models sit first, and the other companies follow in the order the list gives them.

Why is there no search box in the model list?

The list adds the Search models box once it holds more than six models. With a shorter list, every model fits without one.

Which effort levels can I choose?

Off, Minimal, Low, Medium, High, Extra high and Max, as far as the model supports them; a model that offers no levels shows no Effort control. Default appears where a model carries no default of its own.

I changed the effort and the model changed too. Why?

The Effort control belongs to the model highlighted in the list. If you change the level for a model the chat is not using, Oda switches the chat to that model; the new level applies at once.

Can I run the model list with the keyboard?

Yes. The list opens on the chat's model: Up/Down, Home/End and PageUp/PageDown move through the rows, and typing searches. Enter picks the highlighted model and closes the list; Tab moves between the rows and the Effort control, and Left/Right change the level. Escape, or Shift+Tab from the rows, closes back to the model control.

Part of the model list failed to load. What can I do?

Choose Retry where the failure shows. When one company's models cannot be read, its group shows {name} failed to load: {message} and the other groups stay usable; when the whole list cannot be read, the retry sits on the list itself. Rows that cannot be resolved stay dimmed and unselectable.

What does Model operation failed mean?

A model change was refused, and the message carries the reason. Nothing about the chat changes, and the list stays open so you can pick another model or fix what the reason names.

Why did the model control shrink to just an icon?

The control drops its text when the row runs out of space, so it collapses to the model's mark. The full choice stays in its tooltip and in the list, and the name comes back when there is room.

Can I change the model of a subagent's chat?

No. A subagent's chat does not offer the model picker or /model: a subagent takes the decision recorded for the chat that started it. Choosing the model for a subagent covers subagent models.