For a long debugging session, do you start with the strongest model or the cheapest one?

Long debugging cost depends on context resets, review time, and test follow-through, not only the model price.

For a short syntax issue, cheaper tools are usually enough. For a long bug hunt, #node:5620 and #wiki:model-switching-cost-ai-tools-1800 point to a different question: how much context and review time is lost when you switch models mid-session? Do you start with the strongest model for hard bugs, or begin cheap and escalate only when the first pass fails?