The assistant starts with faster, lighter models for speed. Most straightforward actions — creating a record, updating a field, answering a simple question — are handled quickly.

If the results come back unclear, or if you push back on an answer, the assistant automatically escalates to higher-reasoning models for deeper analysis.

This happens behind the scenes. You'll notice that simple actions are fast, and more complex questions get more thorough answers.

You get quick responses for routine tasks without waiting for a heavy model to think.

Complex or ambiguous requests get the reasoning power they need without you asking for it.

The assistant catches its own mistakes — if a lighter model returns something that doesn't look right, it escalates before showing you a bad result.

You don't need to do anything to trigger escalation. Just use the assistant normally.

If you feel like an answer was too superficial, say so — the assistant will re-process with a stronger model.