Choose a Work Mode
Choose a work mode
Before sending your first message, choose whether you need a quick answer or want to complete a piece of work. Select the work mode and response model separately.
| Mode | Best for |
|---|---|
| Quick Q&A (Beta) | Data queries, knowledge questions, simple analysis, and follow-up questions, with faster responses and lower Credit consumption. Explicitly select semantic objects with @ before querying them. Add Knowledge Bases or local files as context as needed. |
| Deep Work | Complex analysis, content generation, Skill invocation, app connections, and business operations. This is the default for new conversations. |
Quick Q&A does not support Skills, app connections, or multistep tasks. Work mode cannot be changed after a conversation is created. To change modes, create a new Chat from a selected response. Digital Worker teams support only Deep Work.
Choose a response model
New Chats offer three response model tiers:
| Tier | Best for |
|---|---|
| Lightweight | Simple questions where response speed matters |
| Standard | Everyday work with a balance of quality and speed |
| Flagship | Complex analysis and reasoning requiring better results |
Switching to Quick Q&A automatically changes a selected Flagship model to Standard; you can still select another tier manually. Switching to Deep Work leaves the model unchanged. Selecting Flagship in Quick Q&A displays a small red dot on the mode entry with the hint "Choose Lightweight or Standard for faster answers."
In addition to these tiers, you can select administrator-configured models directly. These options display model names and suit tasks where you already know which model to use.
The input area shows the current work mode and model tier, such as "Deep Work · Standard". Open the selector to adjust them before sending a message.
Changes apply only to new conversations started afterward. They do not alter existing answers or restart running tasks.
Model tiers affect speed, capabilities, and resource consumption. Choose based on task complexity. If the result is insufficient, add resources or use a higher tier instead of repeatedly retrying for a different answer.