ProductivityBeginner12 min read

Claude 101 (2) - Pick the Right Claude Model: A Plain-English Guide to Saving Tokens

Learn how to choose the right Claude model for each task — so you stop wasting your monthly quota on jobs that didn't need the smartest brain in the room.

Claude 101 (2) - Pick the Right Claude Model: A Plain-English Guide to Saving Tokens

Pick the Right Claude Model: A Plain-English Guide to Saving Tokens

In this guide, you will learn how to choose the right Claude model for each task — so you stop wasting your monthly quota on jobs that didn't need the smartest brain in the room.

Required Tools: Claude Pro | Updated: May 2026

Overview

Claude offers three main models — Opus, Sonnet, and Haiku — and most people just leave it on the default and burn through their message limit by the second week of the month. In this guide, we'll break down what each model is actually good at, give you a one-glance decision table, and show you how to switch models mid-conversation to save tokens without losing context. No jargon, no benchmarks — just real-life tasks and which model to use.

Who This Is Useful For

  • Pro users who keep hitting the message cap before the month is over
  • People who feel guilty asking Claude "small" questions because they think it's wasteful
  • Anyone curious about the difference between Opus, Sonnet, and Haiku but doesn't want a technical paper
  • What You Will Build

    A simple mental model — and a printable cheat sheet — for picking the right Claude model in under 2 seconds. By the end, you'll be able to glance at any task ("translate this email", "plan my trip", "review my contract") and instantly know whether to reach for Haiku, Sonnet, or Opus.

    You'll also learn the one-line trick for switching models mid-chat — so you can start a conversation cheap, and only "level up" to Opus when the question actually deserves it.

    What You Need

  • A Claude Pro, Max, or Team account (Free tier only gives you Sonnet)
  • 5 minutes
  • One real task you've been meaning to ask Claude about — we'll use it to test
  • Step 1: Meet the Three Models

    Forget the version numbers for a second. Here's how to think about the three models like coworkers:

  • Haiku is your fast, cheap intern. Great at simple, well-defined tasks. Doesn't think too deeply, but turns work around in seconds.
  • Sonnet is your reliable senior teammate. Smart enough for 80% of real work, fast enough to feel snappy, and balanced on cost. This is the default for a reason.
  • Opus is your most experienced expert. Slower, more expensive per message, but noticeably better at hard reasoning, long documents, ambiguous problems, and creative writing.

  • Step 2: The 2-Second Decision Table

    Pin this somewhere. When a task lands on your screen, match it to the row and use that model.

    Task TypeBest ModelWhy
    Translate a short messageHaikuFast, simple, no nuance needed
    Summarize an email or articleHaikuPattern matching, not deep thinking
    Reformat data, fix typos, basic editingHaikuBoring jobs done quickly
    Draft a normal email or messageSonnetTone matters, but not life-or-death
    Plan a trip, weekend, or scheduleSonnetMulti-step but predictable
    Brainstorm ideasSonnetVariety beats depth here
    Explain a concept you don't understandSonnetClear teaching, decent depth
    Build something with Artifacts (calculator, mini-app)SonnetCode generation sweet spot
    Review a contract or important documentOpusHigh stakes, nuance matters
    Make a big decision (job offer, major purchase)OpusYou want the best reasoning
    Long creative writing (story, speech, eulogy)OpusTone and depth
    Research that involves weighing tradeoffsOpusGenuine analysis, not summary
    Coding help with tricky bugsOpusWorth the extra cost


    Step 3: Switch Models Mid-Conversation

    You don't have to commit to one model for a whole chat. Above the message box, there's a model dropdown — you can change it at any time, and Claude keeps the full conversation history.

    This is the single biggest token-saver. Real example flow:

    1. Start with Haiku — paste a long article and ask "summarize the key points in 5 bullets"
    2. Switch to Sonnet — "based on those points, draft a LinkedIn post in my voice"
    3. Switch to Opus — "now critique the post and suggest the strongest hook"

    You used the cheap model for grunt work, the mid-tier for the bulk of the writing, and only paid for Opus on the one step where deep judgment actually mattered.


    Step 4: The "Pre-Flight Check" Prompt

    Not sure which model fits your task? Ask Claude itself. Paste this into a new chat:

    
    I have the following task: [describe your task in 1–2 sentences].

    Tell me:
    1. Which Claude model (Haiku, Sonnet, or Opus) is best for this and why
    2. An estimate of how many messages this might take
    3. Whether I should split it into smaller subtasks to save tokens

    Use the cheapest model (Haiku) to run this check — you don't need a genius to recommend a genius. The answer will usually steer you to Sonnet, occasionally to Opus, rarely to Haiku for the actual work.

    Step 5: Three Habits That Cut Quota in Half

    Once you internalize the table, layer on these habits:

  • Default to Sonnet, not Opus. Reset your default model in Settings → Default model. Then escalate manually only when needed.
  • Batch small jobs into one prompt. Instead of 5 separate Haiku chats ("translate this", "translate this", …), paste 5 items in one message: "translate all 5 below, numbered." One message, same result.
  • End long Opus chats and re-summarize in Haiku. When an Opus conversation gets long, ask it to summarize the conclusion. Copy that summary into a fresh Haiku chat for follow-up questions. You keep the insight, drop the cost.

  • Going Further

    Try it on a real task today. Pick something you'd normally throw at the default model. Use the decision table, run it on the recommended model, and see if the answer is good enough.

    Build a "model switcher" Style. Create a Style in your Claude settings called "Cheap Mode" with the instruction: "Default to short answers. Ask before doing heavy reasoning. Suggest if a smaller model would suffice." Switch into it whenever you feel quota anxiety.

    Bring it to your team. If you share a Team plan, this is the single biggest lever for keeping the seat count low. Print the table from Step 2 and pin it in your shared workspace.

    Key Takeaways

    Here's what you learned in this guide:

  • Three models, three roles. Haiku = fast intern, Sonnet = senior teammate, Opus = expert specialist.
  • Default to Sonnet. It handles ~80% of real-world tasks well — Opus is for the 20% that genuinely needs it.
  • Use the 2-second decision table. Match the task type to the row, pick the model, move on.
  • Switch mid-chat to save tokens. Start cheap, escalate only on the steps that need depth.
  • Batch small jobs. Five translations in one message is cheaper than five separate chats.
  • Run a pre-flight check on Haiku. When unsure, ask the cheap model which model to use.
  • You'll feel the difference within a week — quota lasts longer, replies feel more appropriate, and you stop second-guessing whether a question is "worth it" to ask.

    Up next: Article 03 — Master Claude Projects: Give Claude Long-Term Memory of Your Life.

    Learn AI, after work

    Track your progress, earn XP, and unlock more free tutorials in the AfterWork Bytes app.

    Open this tutorial in the app