Academy
7. Overriding: When and How to Choose Manually
Know how to read the model badge, pick a model manually for the rest of a conversation, regenerate with a different model to compare, and switch mid-conversation without losing history.
Every answer in OUPI carries a model badge — a small label just below the response that tells you exactly which model was used. When automatic selection is active, this badge is your window into the routing decision: it confirms whether a frontier, balanced, fast, or specialist model handled your message. Reading the badge is the first step before deciding whether to override. If the badge shows a light model on a task you consider complex, you know it is time to step in.
To pick a model manually, open the model selector in the chat interface. Once you choose a model, that choice holds for the rest of the conversation — every subsequent message will use it until you change again. This is useful when you know the task ahead is hard (legal reasoning, multi-document synthesis, complex code) and you want a frontier model from the start, or when data sensitivity requires a specific provider. You do not need to re-select on every message; one choice locks it in.
Want to compare without committing? Use the regenerate action on the last answer and pick a different model. The same prompt is sent to the new model, so you can see both outputs side by side. This is ideal for borderline tasks — say, a summary that a balanced model handled but you suspect a frontier model would do better. Compare, then decide which model to keep for the rest of the conversation. Your history is never lost when you switch or regenerate.
Use the "balanced by default, frontier for the hard part" rule. Frontier models can cost 10–30× more per token than fast ones. Switch up only when the task demands it — deep reasoning, long multi-step analysis, or when a balanced model visibly struggled. Switch down to a fast model for short, repetitive, low-stakes work.
Switching models mid-conversation does not erase anything. The new model reads the full conversation history and continues seamlessly. This means you can start a conversation with automatic selection for easy turns, then escalate to a frontier model for a difficult question, and even drop back down afterward. Each model picks up where the last one left off. It is also a practical way to test whether a different provider gives a better answer on a specific point.
If your organization restricts which models are available, automatic selection already respects those rules — it only picks from approved models. When you override manually, the selector shows only the models your plan and team settings allow, so you cannot accidentally route data to an unapproved provider.
Open a chat, send a question, and read the model badge on the answer. Then open the model selector, pick a different model, and regenerate the same answer. Compare both outputs. Finally, send a new message — notice it uses your manual choice. Switch back to automatic selection to let OUPI route again.
The model badge tells you what was chosen. Open the model selector to override — your pick holds for the conversation. Regenerate with a different model to compare outputs side by side. Switch models mid-conversation freely; history is preserved. Use frontier models for hard tasks, balanced for everyday work, and fast for simple jobs. Your team's model restrictions apply whether selection is automatic or manual.