Model selection as a judgment call
Match the model to the task instead of always grabbing the top one
After this, you'll be able to explain what Haiku, Sonnet, and Opus are each best at, check which models your account can use, and decide which tasks are worth choosing a lighter or heavier model when the selector is available.
Before you start
Complete Claude Skills: what they are and how to use them first; this lesson builds on the same recurring-task lens, because the model running your work shapes how fast and how cheaply Claude completes what a Skill applies.
The idea
The right Claude model is the one that fits the task, not the most powerful one you can reach for. Claude comes in three tiers: Haiku for fast, cheap, routine work, Sonnet for most work, and Opus for the hardest reasoning.
The exact selector you see depends on your plan and the surface you are using. That makes the current account menu the source of truth, while the judgment stays the same: match the model to the task.
Here is the before and after: On Pro without thinking, you run "reword this one sentence" on the heaviest model, which is slower and spends more usage than the task needs. With model choice as a habit, you send quick rewords to Haiku, keep drafting and analysis on Sonnet, and save Opus for the 40-page contract.
Now try it take a quick reword and a normal draft from your day, then decide which model each would suit. If your account exposes a selector, choose intentionally; if it does not, build the judgment for the surfaces where it does.
Match the model to the task, and stop spending heavy-model time on Haiku jobs.
Try it (10 min)
Watch out for
- On Pro, defaulting to the heaviest model for everything, which slows your quick tasks and spends usage you did not need to spend
- Using Haiku for genuinely hard reasoning and then trusting an answer that needed more depth than it got
- Assuming every plan and surface exposes the same model menu; check the current selector before you decide
- Picking a model by name recognition instead of by task; the question is always what this task needs, not which sounds best
- Switching models mid-conversation and expecting the new model to have full memory of the chat without you re-stating what matters
Paste this into Claude
I want to get better at choosing the right Claude model instead of always using the most powerful one. I know model availability depends on my plan and the Claude surface I am using, so I am building the judgment for when I can switch. Here are three real tasks from my work this week: TASK A (quick and routine): [e.g. "fix the grammar in this two-line message"] TASK B (normal everyday work): [e.g. "draft a first version of a project update email"] TASK C (genuinely complex): [e.g. "read this 30-page vendor contract and flag anything risky"] For each task, tell me: 1. Which model fits (Haiku, Sonnet, or Opus) and, in one plain sentence, why 2. For Task C, whether my available model could already do a decent job, or whether this is genuinely worth switching to Opus if my account supports it 3. What I would waste if I used a heavier model than a task needs Then give me a simple rule of thumb I can use whenever my account exposes model choice, and remind me to trust the current selector over old plan notes.
What good looks like
- Claude matched Haiku to the routine task and Sonnet to the everyday task, and reminded you to check the current selector
- Claude treated Opus as the deeper option for the complex task and said whether your available model could still handle it
- Claude named the concrete waste of going too heavy for the routine task
- Claude gave you a rule of thumb short enough to remember
- You can now look at a new task and name the model that fits in under five seconds, ready for when you can switch
When this breaks
- Breaks when you treat model choice as a quality setting (heavier is always better) because the most capable model is slower, so on simple tasks you trade speed and usage for depth the task cannot use.
- Degrades when you never revisit your model on Pro because one becomes your reflex; routine work crawls on a heavy model, or hard work runs thin on a light one, and you stop noticing either.
AI can help with this
Paste this: 'Here are three tasks: A is quick, B is everyday, C is complex. Match A to Haiku and B to Sonnet, tell me whether C is worth Opus if my account supports it, and give me a one-line rule of thumb for choosing from the models my current selector shows.'
You can now
You can complete the lesson outcome in a real Claude chat, Project, Artifact, Connector, Desktop, or Code surface.
- You can verify that claude matched Haiku to the routine task and Sonnet to the everyday task, and reminded you to check the current selector.
- You can verify that claude treated Opus as the deeper option for the complex task and said whether your available model could still handle it.
- You can verify that claude named the concrete waste of going too heavy for the routine task.
- You can verify that claude gave you a rule of thumb short enough to remember.
Key takeaways
Choosing a Claude model is a judgment call about task depth, not a quality dial. Check the current selector for what your account supports, then reach for Haiku or Opus based on what each task actually needs.
- Sonnet handles most everyday work.
- Haiku suits quick, routine jobs where speed matters more than depth.
- Opus suits long documents or reasoning that genuinely needs it.
- Model availability varies by plan and surface, so check the current selector before relying on old notes.