Claude is not one model. It is a family of models with different sizes, speeds, and price points, and learning when to reach for each is a craft skill that pays for itself many times over. This chapter is a guided tour.
3.1 The model lineup at a glance
At the time of writing, Anthropic ships three model tiers:
| Tier | Strength | Best for |
|---|---|---|
| Opus | Deepest reasoning, broadest knowledge | Hard problems, big refactors, novel architecture |
| Sonnet | Balanced quality, speed, and cost | Daily-driver coding, the default for almost everything |
| Haiku | Fastest, cheapest | High-volume tasks, simple classifications, latency-sensitive UI |
Within each tier there are versions (Sonnet 4.6, Opus 4.7, and so on). Newer versions are usually better at the same price. When in doubt, use the latest of each tier.
3.2 Opus — when you need the deepest thinking
Opus is the model you reach for when the problem is genuinely hard: you need to reason across many files, design a non-obvious algorithm, or untangle a subtle bug that has resisted three rounds of fixes. It is the slowest and most expensive of the three, often by an order of magnitude, so use it deliberately. A useful heuristic: if you would have asked your most experienced colleague for help, that is an Opus moment.
3.3 Sonnet — the daily-driver workhorse
Sonnet is what you should default to. It handles the overwhelming majority of coding tasks with quality close to Opus and at a fraction of the cost. If you are not sure which model to pick, pick Sonnet. The whole TaskFlow capstone in Part IV is built on Sonnet, with Opus reserved for two specific moments.
3.4 Haiku — when speed and cost matter most
Haiku shines in places where you are calling the API many times, where the task is well-defined, and where the user is waiting on the answer. Examples: classifying support tickets, summarising a single paragraph, generating a quick caption. It is dramatically cheaper than Sonnet and several times faster, but its reasoning is shallower. Do not ask Haiku to design your database.
3.5 Context windows
The context window is how much text Claude can read in a single conversation — both your input and its output. Modern Claude models have very large windows (hundreds of thousands of tokens, the equivalent of a medium-sized book). In practice this means you can paste an entire small codebase into a chat without splitting it.
Two practical implications. First: do not be stingy about context. If pasting more code helps, paste more code. Second: the window is not infinite, and as the conversation grows Claude's attention spreads thinner. A focused 10,000-token conversation will out-perform an unfocused 200,000-token one.
3.6 Tokens, pricing, and back-of-envelope math
A token is roughly four characters of English, or about three-quarters of a word. A page of dense text is around 500 tokens. The Anthropic API charges per million tokens, with separate prices for input and output. Output is more expensive than input.
The TaskFlow "smart suggestions" feature you build in Chapter 21, for example, sends roughly 800 tokens of input and gets back around 300 tokens of output per call. At Sonnet pricing that is well under a cent per call. Even a thousand calls a day costs less than a coffee.
3.7 Switching models
You change models in three places, depending on which surface you are using.
Claude.ai. The model picker is at the top of the chat window. Switching mid-conversation works; the new model picks up where the old one left off.
Claude Code. The CLI defaults to Sonnet. Override with --model:
claude --model claude-opus-4-7
# or, inside an existing session:
/model claude-opus-4-7
The API. The model name is a parameter on every request:
const message = await anthropic.messages.create({
model: "claude-sonnet-4-6",
max_tokens: 1024,
messages: [{ role: "user", content: "Hello, Claude." }]
});
3.8 A decision flowchart
- Is the task simple, well-defined, and either high-volume or latency-sensitive? → Haiku.
- Is the task hard enough that you would ask a senior colleague? → Opus.
- Otherwise → Sonnet.
This flowchart fits on a sticky note. Put it on your monitor for the first month. After that, the choice will become automatic.
Key Takeaway. Default to Sonnet. Escalate to Opus when the problem genuinely demands it. Drop to Haiku when speed or cost matters more than depth. Model choice is a craft skill — practice it on day one, and it becomes second nature by the end of this book.