Claude gives you several models to choose from. Their names—Haiku, Sonnet, Opus, and now Fable—do not tell a new user much about which one should write an email, analyze a spreadsheet, or help plan a project.
Here is the practical answer:
Start with Claude Sonnet 5. Use it for almost everything. Switch models only when you have a clear reason.
Sonnet 5 is Anthropic’s current default model for Free and Pro plans. It is strong enough for most writing, research, analysis, planning, and everyday business work. You do not need to study every model before you can start getting value from Claude.
This guide explains the differences in plain English, using the models available as of July 2026. Claude changes quickly, and the models shown in your account may vary by plan, product, and company settings. You can see your current model beside the message box and use the model selector to change it. Anthropic keeps its model-selection instructions here.
The short version
| Model | Think of it as | Use it for |
|---|---|---|
| Claude Haiku 4.5 | Fast assistant | Simple summaries, extraction, classification, quick drafts, and repetitive high-volume work |
| Claude Sonnet 5 | Everyday expert | Emails, proposals, research, marketing, document analysis, planning, spreadsheets, and most business work |
| Claude Opus 4.8 | Senior specialist | Difficult analysis, long projects, important documents, complicated tradeoffs, and work that needs stronger judgment |
| Claude Fable 5 | Special-project team | The hardest, longest, multi-stage projects where extra capability justifies additional usage or cost |
If you remember only one rule, remember this:
Use Sonnet first. Move up when the task is unusually difficult. Move down when speed and volume matter more than depth.
Claude Sonnet 5: the right default for most businesses
Sonnet is the middle of Claude’s traditional model family, but “middle” does not mean average. Anthropic made Sonnet 5 the default for Free and Pro users because it offers a strong balance of capability, speed, and usage.
Use Sonnet 5 for:
- Writing and revising emails
- Turning meeting notes into action items
- Drafting proposals, policies, and standard operating procedures
- Researching customers, competitors, or markets
- Summarizing long documents
- Reviewing a contract before sending it to your lawyer
- Analyzing a spreadsheet and explaining what changed
- Planning a marketing campaign
- Brainstorming and improving an offer
- Creating a presentation or report
- Building a repeatable workflow inside a Project or Skill
For example, suppose you upload three months of customer feedback and ask Claude to identify common complaints, group them by theme, and recommend five changes. Start with Sonnet. It can handle the files, the analysis, and the writing without immediately reaching for the most expensive model.
When Sonnet is not enough
Consider switching if:
- Claude keeps missing an important constraint after you clarify it.
- The work involves many documents with conflicting information.
- You need Claude to plan and complete a long sequence of dependent steps.
- A weak answer would cost substantial time or money.
- You want a serious second pass on work Sonnet already produced.
Before switching, improve the prompt. Give Claude the goal, relevant background, source material, constraints, and the format you want. A better prompt often matters more than a bigger model.
Claude Haiku 4.5: use it when speed matters most
Haiku is the smaller, faster model. Anthropic describes Haiku 4.5 as useful for responsive work such as customer service, real-time assistance, and high-volume tasks.
For a small business, Haiku makes sense when the job is clear, repeatable, and easy to check.
Good Haiku tasks include:
- Pulling names, dates, totals, or categories from documents
- Summarizing a short email thread
- Reformatting text into a standard template
- Tagging support messages by topic
- Producing several simple headline variations
- Translating straightforward internal content
- Drafting basic replies that a person will review
- Handling a repetitive task inside an automation
Imagine that you receive 200 survey responses. You want each response labeled as praise, complaint, feature request, or question. That is a good Haiku job: the categories are defined, the work repeats, and you can spot-check the results.
Haiku is less appropriate when the task is ambiguous, politically sensitive, strategically important, or dependent on subtle judgment. It may still produce a confident answer, but speed is not the same as careful reasoning.
You may not see Haiku in every version of the Claude app. Model availability varies, and businesses using Claude through other tools may encounter Haiku more often because its speed and cost are useful in automations.
Claude Opus 4.8: use it for difficult work that needs judgment
Opus is built for more demanding work. Anthropic positions Opus 4.8 for complex professional tasks, long-running projects, AI agents, and work that benefits from stronger consistency and judgment.
Use Opus 4.8 when you need Claude to:
- Compare several strategic options with competing tradeoffs
- Analyze a large collection of dense documents
- Find weak assumptions in a business plan
- Review an important proposal for gaps and contradictions
- Maintain context across a long, multi-stage project
- Produce polished work that combines analysis, writing, and detailed instructions
- Take a serious second look at a consequential Sonnet answer
For example, suppose you are deciding whether to hire an employee, use a contractor, or automate part of a role. You provide financial assumptions, workload data, risks, and growth plans. Opus is a reasonable choice because the task involves uncertainty, tradeoffs, and decisions that affect real people and real money.
Opus should not become a substitute for professional review. A more capable model can still invent facts, misunderstand a document, or make a bad recommendation. Use qualified legal, financial, medical, or human-resources professionals for high-stakes decisions.
A useful Opus workflow
You do not always need to start over with Opus. Try this:
- Complete the first draft in Sonnet.
- Switch to Opus in the same conversation.
- Ask it to identify unsupported claims, missing considerations, and risks.
- Ask for a revised version that addresses those findings.
- Verify important facts against the original sources.
This gives you more value from Opus than using it for every routine message.
Claude Fable 5: reserve it for unusually ambitious projects
Fable is a newer class of Claude model. It sits above the normal everyday choices and is designed for exceptionally difficult, long-running work. Anthropic describes Fable 5 as a model for ambitious coding and professional projects that may continue through many stages with less supervision.
For a nontechnical business user, possible Fable projects include:
- Synthesizing a large set of reports, spreadsheets, interviews, and research into a detailed recommendation
- Planning and producing a substantial operating manual across many departments
- Auditing a complicated workflow with many tools, handoffs, and failure points
- Building a complete decision package that includes analysis, scenarios, risks, and presentation materials
- Completing a long research project that requires repeated searching, comparison, and validation
Most small-business tasks do not need Fable. An email does not become better because the most powerful model wrote it. A simple summary does not need days of autonomous work. Start with Sonnet and treat Fable as an escalation path, not the default.
Fable access may require a qualifying plan or usage credits. Its availability has also changed since launch, so check what your model selector and plan currently offer.
What about older Claude models?
You may see names such as Sonnet 4.6, Opus 4.7, Opus 4.6, or Haiku 4.5 under “More models,” inside Claude Code, or in another product that connects to Claude.
Older models remain available for reasons such as:
- A company tested a workflow on one version and wants consistent behavior.
- A connected application has not moved to the newest model.
- A developer needs a specific balance of price, speed, and capability.
- A team needs time to evaluate a new release before changing its standard model.
If you are new to Claude, you generally do not need to choose an older model. Use Anthropic’s recommended default unless a trusted workflow or company policy tells you otherwise.
You may also hear about Mythos 5. It is a restricted research configuration rather than a normal choice for everyday Claude users. You can ignore it when deciding which model should draft your next proposal or analyze customer feedback.
Model, effort, and extended thinking are different settings
Claude’s menu can include three separate choices:
- Model: Which version of Claude handles the task.
- Effort: How much work the selected model puts into its response.
- Extended thinking: Whether Claude spends additional time reasoning before answering.
These controls can look like more models, but they are settings applied to a model.
Use more effort or extended thinking when:
- The question has several valid answers and meaningful tradeoffs.
- Claude must plan before it writes.
- You want it to check assumptions and consider failure cases.
- The task is expensive to redo.
Leave them off for quick rewrites, simple summaries, and routine questions. More thinking can improve difficult answers, but it also takes longer and may use more of your plan’s allowance.
Which model should you use for common business tasks?
| Business task | Start with | Switch when |
|---|---|---|
| Rewrite an email | Sonnet | Haiku if you need many simple variations quickly |
| Summarize meeting notes | Sonnet | Haiku for a clear, repeatable summary format |
| Draft a proposal | Sonnet | Opus for a high-value proposal with complicated requirements |
| Analyze customer feedback | Sonnet | Haiku for simple tagging; Opus for subtle themes across many sources |
| Review a spreadsheet | Sonnet | Opus when several sheets, assumptions, and scenarios interact |
| Create a marketing plan | Sonnet | Opus when channels, budgets, positioning, and forecasts must agree |
| Write an SOP | Sonnet | Opus for a large process with many roles and exceptions |
| Make a major business decision | Sonnet with thinking | Opus or Fable for deep research and competing scenarios |
| Run a repetitive automation | Haiku or Sonnet | Use Sonnet when mistakes require judgment to detect |
| Complete a multi-day research project | Opus | Fable when the scope is unusually large and access justifies it |
A simple decision process
When starting a new Claude conversation, ask these questions:
1. Is this a normal business task?
Use Sonnet.
2. Is it simple, repetitive, and easy to verify?
Consider Haiku if it is available.
3. Is it difficult, consequential, or full of competing constraints?
Try Opus, especially as a reviewer after a Sonnet draft.
4. Is it an unusually large project that may take many stages?
Consider Fable if your plan includes it and the extra usage is justified.
5. Are you disappointed with the answer?
Do not switch models immediately. First add context, examples, constraints, and a clear definition of a good result. If the better prompt still fails, move up.
Main prompt: choose the right model before doing the work
Run this in your normal Claude model. It will help you determine whether the task needs more speed, more judgment, or simply a better brief.
Evaluate this business task before completing it.
TASK
[DESCRIBE THE OUTCOME YOU NEED]
CONTEXT AND SOURCES
[PASTE OR ATTACH RELEVANT MATERIAL]
CONSTRAINTS
- Audience: [AUDIENCE]
- Deadline: [DEADLINE]
- Required format: [FORMAT]
- Cost of an incorrect result: [LOW, MEDIUM, OR HIGH — EXPLAIN]
- Human review available: [DESCRIBE]
FIRST, ASSESS THE TASK
1. Rate its complexity, ambiguity, consequence, and ease of verification
from 1–5.
2. Recommend the appropriate Claude model tier: fast/lightweight,
balanced/default, or highest-capability. Use the current model names
available in my account rather than assuming a specific name.
3. Recommend whether additional thinking or research is justified.
4. Explain the choice in plain English.
THEN
- If the current model is appropriate, complete the task.
- If a different model is materially better, stop after creating a clean
handoff brief I can paste into that model.
FINISH WITH
1. The deliverable or handoff brief.
2. Assumptions and unsupported claims.
3. A human-review checklist proportional to the risk.This avoids memorizing model names that may change. You are choosing based on the work: speed, complexity, consequence, and how easily a human can check the result.
The model matters less than the working relationship
People often spend too much time hunting for the “best” model and too little time explaining the work.
Claude will usually perform better when you provide:
- The outcome you need
- The audience for the work
- Relevant files and background
- Examples of good and bad results
- Constraints it must respect
- A clear format for the response
- Feedback on what needs to change
The best model cannot guess the facts that only you know. A clear brief, useful source material, and a few rounds of feedback will often beat switching models repeatedly.
Start with Sonnet 5. Give it a real piece of work. Move to Haiku for speed, Opus for judgment, or Fable for exceptional scope. That is enough model strategy for most businesses getting started with Claude.