How quickly ZeroTwo answers depends mostly on the model you picked, how much it reasons before answering, and which tools it uses. A short chat message to a fast model comes back quickly. A Work or Code task that takes many steps does not.
Pick a faster model
Open the model picker in the chat and try Auto. Auto is ZeroTwo's own choice that routes your message to a fast, low-cost model. In Chat, Work and Code it routes to GPT 6 Luna. On paid plans, Auto requests come out of your free Auto allowance first.
Lower the thinking effort
For models that can reason before they answer, the message box has a Thinking effort control. In Chat it offers None (for models that can turn thinking off), Standard and Extended. In Work and Code it lists the model's own levels. Choose a lower level for less reasoning before the answer.
Tasks with many steps
Web search, Deep Research, Work and Code do more than generate text. In Work, ZeroTwo takes a task through several steps with tools and shows the steps as it goes, so you can see that it is still working. While it runs, Settings > General > Follow-up behavior lets you choose whether a new message waits in line (Queue) or redirects the current run (Steer).
If the connection drops during a task
In an agent run such as Work or Code, ZeroTwo tries to reconnect on its own and shows Reconnecting… with a count out of 5. If it gives up, a Connection interrupted row appears with a Retry button. If a tool was still running when the connection dropped, ZeroTwo asks whether it already ran: choose It already ran or Run it again. Pick carefully if the tool sends messages or changes files.
More things to try
- Check that your internet connection works, then reload the page or restart the app and send the message again.
- If it keeps happening, report it. Select the question-mark button (Help and what's new), choose Help > Report Bug, and say which model and mode you used and what the message was.