Most people use one AI model for everything. They open ChatGPT or Claude, type their prompt, and accept whatever model is set as default.
That works for simple tasks. It wastes time and money for everything else.
Different models have different strengths. Using the right one for the right task saves time, reduces cost, and produces better output. Here is a practical guide to matching models to tasks.
The Models (Claude Family, June 2026)
Haiku
Fast. Cheap. Good enough for simple tasks. Not built for complex reasoning.
Use for: Quick summaries, simple formatting, basic Q&A, short email replies, extracting data from structured text.
Do not use for: Strategy, complex analysis, nuanced writing, anything where you need depth.
Sonnet
The daily workhorse. Solid quality for most tasks without the cost or latency of the bigger models.
Use for: Code reviews, email drafting, report generation, meeting notes, blog outlines, workflow templates, most professional writing.
Do not use for: Deep strategic analysis where you need the model to hold multiple competing factors in its head at once.
Opus
The deep thinker. Best for complex reasoning, strategy, and decisions with many variables.
Use for: Strategic decisions, evaluating business options, multi-criteria analysis, complex code architecture, anything where quality matters more than speed.
Do not use for: Simple tasks where Sonnet would produce the same quality at lower cost and faster speed.
Fable
The newest model. Built for complex reasoning and tough problems. One tier above Opus in analytical depth.
Use for: The hardest problems. When Opus gives you good output but you sense it could go deeper. Complex strategic decisions with high stakes.
Do not use for: Anything routine. You are paying for capacity you do not need.
The Effort Setting
Most AI tools have an "effort" or "thinking" setting. This controls how long the model spends reasoning before generating its response. It matters as much as model choice.
Low effort: Fast output, lower quality. Good for when you need a quick draft and will edit it yourself. Email replies, quick summaries, first passes.
Medium effort: Balanced. The default for most tasks. Good quality without excessive thinking time.
Max effort: Slowest output, highest quality. The model thinks deeply before responding. Good for strategic analysis, complex decisions, and anything where you want the best possible answer for the context you provided.
The relationship between model and effort: a Sonnet response at max effort can sometimes match an Opus response at low effort. The difference is not just the model. It is how hard you are asking it to think.
A Practical Matching Guide
| Task | Model | Effort |
|---|---|---|
| Summarize a document | Sonnet | Low |
| Write a professional email | Sonnet | Medium |
| Code review | Sonnet | Medium |
| Generate a workflow template | Sonnet | Medium |
| Blog post draft | Opus | Medium |
| Evaluate career options | Opus | Max |
| Business strategy analysis | Opus | Max |
| Complex product decision | Fable | Max |
| Multi-variable tradeoff analysis | Opus or Fable | Max |
| Quick data extraction | Haiku | Low |
| Format and clean text | Haiku | Low |
The Cost Question
Using Opus for everything is not "better." It is wasteful. You burn through tokens on tasks where Sonnet would produce identical output at a fraction of the cost. The money you save on the simple tasks is money you can spend on the complex ones, where the bigger model actually makes a difference.
Think of it like tools in a workshop. You do not use a table saw to cut a piece of sandpaper. You use scissors. Save the table saw for the job that actually needs it.
When to Switch Mid-Task
Sometimes you start with one model and realize you need another. That is fine. If Sonnet gives you a surface-level analysis and you need depth, bump it to Opus with a follow-up prompt: "Go deeper on the tradeoffs between options 2 and 3. Consider second-order effects."
If Opus is taking too long on a simple formatting task, drop it to Sonnet. Adapt as you go.
The Bottom Line
Model selection is not complicated. It comes down to one question: how much does quality matter for this specific task? Use the smallest model that can deliver the quality you need. Save the big models for the decisions that deserve them.