Choosing the right AI model for the job

Start with the question, not the model

The instinct is to always reach for the “best” model. In practice, the right question is what the task actually needs — speed, cost, and reasoning depth trade off against each other, and most work doesn't need the most expensive option.

Don't guess — compare

For anything where the answer actually matters, use Compare modelsin the composer instead of picking one and hoping. Send the same prompt to two or three models, see the responses side by side, and let whoever's in the session weigh in on which one to go with. The choice gets recorded, so the reasoning behind it doesn't disappear once the conversation moves on.

Cost adds up faster than people expect

A heavier model isn't just slower — it costs meaningfully more in credits per response, since output tokens especially are priced higher on frontier models. If your team defaults to the priciest model for everything, you'll burn through a monthly credit allotment far faster than expected. Switching to a lighter model for routine questions and saving the expensive one for what actually needs it is the single biggest lever for stretching your budget.

A simple default that works for most teams

• Set your workspace's default model to a solid mid-tier option — not the cheapest, not the most expensive
• Reach for a top-tier model deliberately, when the stakes justify it
• Use comparison mode for genuinely close calls, not as a default habit

For the mechanics of switching models and reading credit costs, see Comparing AI models side by side and How credits work.