Higher tier usually means better judgment and harder reasoning, but also more cost and more waiting, so the skill is matching the model to the job.
Think about choosing a vehicle for a delivery. A box truck carries a couch across town, but it’s slow and overkill for a one-page envelope. A bike courier is cheaper and faster for the small delivery and wrong for the couch. A top-tier model is the box truck for heavy judgment: strategy, architecture, hard writing, or debugging where the answer isn’t obvious. A cheaper model is the courier for repeatable work once the instructions are clear.
How it shows up
When you’re learning a workflow, building a first version, or reasoning through a mess, start with the strongest model you have; we default there until limits or cost bite. Once the work is understood, the choice changes: if the strong model helped you build a clear skill or template, you can often run that same work on a faster one. That’s the Opus-then-Haiku pattern we teach: the stronger model figures out the job, the cheaper model repeats it once the path is documented.
You’ll feel tiers inside Claude Code: the agent may ask which model to use, and feel slow on a harder one, snappy on a lighter one. Tier also affects inference, the model run itself: a stronger model spends more time thinking, a lighter one answers fast but misses edge cases.
Why you care
Names and rankings change; the matching problem stays. Stop sending the box truck for every envelope.