Free guide for Maryland small businesses

Which Claude Model Should You Use?

Claude comes in four models: Haiku, Sonnet, Opus and Fable. They trade speed for depth. Picking the right one for the job gets you better answers, faster, without burning through your plan's limits. This guide explains what each is good at, where each falls short, and which to use when.

An independent guide from Crimson Business Consulting. Not affiliated with, sponsored by, or endorsed by Anthropic.

The basics

What's the difference between models?

A model is the engine behind Claude. All four current models come from the same company and share the same basic skills: they read your files and images, write in many languages, and follow instructions. The difference is how hard each one thinks before it answers.

Think of them as four hires at different levels. You wouldn't send your most senior person to answer a simple phone call, and you wouldn't give your newest assistant the lease negotiation. Same idea here.

Speed
Smaller models answer in seconds. Bigger ones think longer before they reply.
Depth
Bigger models catch more, handle longer multi-step work, and make fewer mistakes on hard problems.
Cost
Bigger models use more of your plan's limits, and cost more per word if you pay by usage.
Faster, cheaperDeeper, slower
HaikuThe fast assistant. Quick, simple, cheap.
SonnetThe reliable staffer. Handles most everyday work.
OpusThe senior pro. For work that has to be right.
FableThe outside specialist. For the rare, hardest jobs.

Meet the models

What each one is good at, and where it falls short

Versions current as of October 2026. "Relative price" compares Anthropic's per-word pricing for developers, with Haiku as 1x. On a subscription plan, the same gap shows up as how fast you reach your usage limits.

Haiku

Haiku 4.5

The fast assistant

Speed FastestRelative price 1xKnows events to Feb 2025

Reach for it when

  • You need a quick rewrite, caption or one-line answer
  • You're sorting, tagging or cleaning up a long list
  • Speed matters more than polish

Strengths

  • Fastest and cheapest model
  • Can still think through a problem when asked
  • Barely dents your usage limits

Watch out for

  • Its knowledge stops in early 2025, so turn on web search for anything recent
  • Smaller working memory than the others
  • Misses more on hard reasoning
  • No effort setting to turn up

Plans: Free and all paid plans.

Sonnet

Sonnet 5.5

The reliable staffer

Speed FastRelative price 2xKnows events to Jun 2026

Reach for it when

  • You're drafting emails, posts or customer replies
  • You want to summarize a document or meeting notes
  • You're brainstorming and want fast back-and-forth

Strengths

  • Anthropic's best balance of speed and intelligence
  • Good at writing, everyday analysis and reading images
  • Large working memory, same as Opus

Watch out for

  • Less thorough than Opus on long, multi-step work
  • Can miss subtle issues in contracts or numbers

Plans: Free and all paid plans. The best choice on the free plan.

Opus

Opus 5.5

The senior pro

Speed ModerateRelative price 4xKnows events to Jun 2026

Reach for it when

  • You're reviewing a contract, lease or policy
  • You're analyzing a spreadsheet or sales data
  • You need research or a memo you'll act on

Strengths

  • Anthropic's recommended starting point for most serious work
  • Built for long, multi-step knowledge work
  • Reads charts, diagrams and screenshots precisely
  • Effort setting lets you trade speed for thoroughness

Watch out for

  • Slower than Sonnet, since it always thinks first
  • Uses your plan's limits faster
  • Not available on the free plan

Plans: Pro, Max, Team and Enterprise.

Fable

Fable 5.1

The outside specialist

Speed SlowestRelative price 10xKnows events to Jun 2026

Reach for it when

  • You need deep, multi-step research with follow-ups
  • You want a finished report, spreadsheet or deck built from scratch
  • You're working through dense filings or long documents

Strengths

  • The most capable model Anthropic offers to everyone
  • Strongest on long research and complete deliverables
  • Best at connecting details across very long documents

Watch out for

  • Slowest and most expensive
  • Summaries can repeat source wording without quotation marks, so check quotes before you reuse them
  • Writing can be dense
  • Overkill for everyday tasks

Plans: Paid plans only. On Max and premium Team or Enterprise seats, up to half your weekly limit can go to Fable at no extra cost. On Pro and standard seats, it runs on pay-as-you-go usage credits.

Interactive

Find the right model for your task

Answer three questions. The recommendation updates as you go.

1. Which plan are you on?

2. What are you trying to do?

3. How much does getting it exactly right matter?

Recommendation

Beyond the model

Settings that matter as much as the model

You can change the model and these settings at any point in a conversation, from the menu next to the send button.

Effort

Sonnet, Opus and Fable let you choose how long Claude thinks. Turning up effort is often a better move than switching to a bigger model.

  • Low, MediumRoutine tasks. Your usage goes further.
  • HighAnthropic's best overall balance of quality and speed.
  • Extra highLong, multi-step tasks, on newer Opus models.
  • MaxThe most thorough. Slowest, and uses the most of your limits.

Your context

A smaller model with your price list, policies and examples will usually beat a bigger model guessing. Put your standing business facts in a Project so every chat starts with them.

The more Claude knows about the audience, the goal and what "done" looks like, the less the model choice matters.

Web search

Every model's knowledge stops at a cutoff date: June 2026 for Sonnet, Opus and Fable, and February 2025 for Haiku. For anything current, like prices, rules, news or a competitor's latest move, turn on web search so Claude looks it up instead of answering from memory.

Side by side

The four models compared

Haiku Sonnet Opus Fable
Best forQuick rewrites, sorting listsEmails, drafts, summariesContracts, numbers, researchDeep research, full deliverables
SpeedFastestFastModerateSlowest
Relative price1x2x4x10x
Working memoryAbout 150,000 wordsAbout 750,000 wordsAbout 750,000 wordsAbout 750,000 words
Knows events up toFeb 2025Jun 2026Jun 2026Jun 2026
Effort settingNoYesYesYes
Free planIncludedIncludedNot availableNot available
Pro planIncludedIncludedIncludedUsage credits
Max planIncludedIncludedIncludedUp to 50% of weekly limit

Working memory (the context window) is how much Claude can hold in mind at once: your files plus the conversation. Word counts are rough conversions of 200,000 and 1 million tokens, at about three-quarters of a word per token. "Included" means within your plan's usage limits. On Team and Enterprise, Fable access depends on whether your seat is standard or premium.

Avoid these

Four common mistakes

Using the biggest model for everything

A customer thank-you note doesn't need Fable. You'll wait longer and hit your limits by Wednesday.

Using the smallest model for what matters

Haiku reading your lease is a false economy. Money, legal and customer-facing work deserves Opus.

Trusting it on recent events

No model knows what happened after its cutoff. Turn on web search, and check dates on anything it cites.

Switching models before adding context

If an answer is thin, first give it more: the file, the audience, an example. Then turn up effort. Then switch models.

For developers: building on the Claude API

If you or a vendor is building Claude into your own software, model choice is also a line item. Prices are per million tokens, in U.S. dollars.

Haiku 4.5Sonnet 5.5Opus 5.5Fable 5.1
Model IDclaude-haiku-4-5-20251001claude-sonnet-5-5claude-opus-5-5claude-fable-5-1
Input / output$1 / $5$2 / $10$4 / $20$10 / $50
Batch (50% off)$0.50 / $2.50$1 / $5$2 / $10$5 / $25
Context / max output200K / 64K1M / 128K1M / 128K1M / 128K
Default effortNonehighmediumhigh
  • Anthropic recommends starting on Opus 5.5, tuning effort, and moving to Fable 5.1 only if your tests at high effort still fall short.
  • Sonnet 5.5, Opus 5.5 and Fable 5.1 reject forced tool use (tool_choice of any or tool). Use strict tool definitions or structured outputs instead.
  • Fable 5.1 carries 30-day data retention and isn't available under zero data retention unless Anthropic authorizes it.
  • Batch processing halves the price for work that can wait. Prompt caching cuts the cost of re-reading the same long context.

Put it to work

Know which model to use? Now give it a job.

Our free Five Jobs tool walks you through five practical tasks, with prompts written for your business. Or talk with us about fitting AI into how your business actually runs.

Try the Five Jobs tool Book a Foundations Session