Which Claude model should you use? Haiku vs Sonnet vs Opus vs Fable

People talk about “using Claude” as if it’s one thing. It isn’t — it’s a family of models that share the same training lineage and values but differ a lot in capability, speed, and cost. And one of the quietest, highest-leverage habits you can build is simply choosing the right one for the task in front of you.

Get it right and you’re fast and cheap on easy work and genuinely powerful on hard work. Get it wrong and you’re either overspending on trivial tasks or stuck because you’ve pointed a sprinter at a marathon.

The four tiers

Think of it like matching a person to a job.

Model	Speed	Cost	Best for
Haiku 4.5	Fastest	Lowest	Quick edits, lookups, simple one-liners, high-volume tasks
Sonnet 4.6	Balanced	Mid	Most coding, writing, and explaining — the everyday default
Opus 4.8	Deep	High	Hard architecture calls, subtle multi-file bugs, deep reasoning
Fable 5	Deepest	Highest	The hardest problems — state-of-the-art across nearly every benchmark

The names map onto an intuition: a haiku is short and quick, a sonnet is the well-rounded standard, an opus is a deep and serious work — and Fable, released in June 2026, is the most capable Claude available, built on the same research as Anthropic’s frontier work.

The one rule that matters

Match the brain to the difficulty of the task. That’s it. Everything else is detail.

• Haiku when you already know what you want and just need execution — rename this across the file, generate this boilerplate, look up this value. If you’d have been happy with an autocomplete, Haiku is the right level of engagement.
• Sonnet for the broad middle: most coding, most writing, most “explain this” and “draft that.” It’s the default for a reason and handles the vast majority of work well.
• Opus when a problem has real depth — a tricky design decision, a bug that spans several files, reasoning that needs to hold a lot in mind at once. When Sonnet has tried twice and keeps missing, this is your move.
• Fable for the genuinely hardest problems, where you want the most capable model there is and cost is the least of your concerns.

How to use this day to day

You don’t decide once and forget it. The good habit is to notice the signal and switch:

Stuck after two tries? Go up one tier — Sonnet → Opus → Fable. The extra cost on a single hard task is almost always worth it, and staying on a model that keeps missing is the expensive option once you count your time.

Churning through quick tasks? Go down to Haiku. Running Sonnet on something Haiku could nail is just quiet waste, multiplied across every call.

Genuinely unsure? Stay on Sonnet. The default is well-chosen, and second-guessing it on ordinary work isn’t worth the friction.

In Claude Code, switching is one command:

/model

to open the picker, or jump straight to one:

/model claude-opus-4-8
/model claude-haiku-4-5

The switch is mid-session — your conversation continues, only the model behind it changes. So a natural rhythm is to cruise on Sonnet, drop to Haiku for a burst of small edits, and bump up to Opus or Fable the moment you hit something thorny.

Pair effort with the model on hard problems

For the toughest tasks, the model is only half the lever. Claude can also spend more effort — more extended thinking before it answers. On a hard problem, pairing a top-tier model with a higher effort or thinking level is how you get the best Claude can do:

/model claude-fable-5
/effort xhigh

For everyday work, leave effort where it is — extra thinking on a simple task is just slower for no gain. Reserve the “top model + high effort” combination for the problems that actually deserve it.

The mistake almost everyone makes

The single most common error isn’t picking the wrong model — it’s never switching at all. People land on one and stay there forever, which means they’re simultaneously overspending on easy tasks and under-powered on hard ones.

Five seconds of “is this task big or small?” before you start, and a willingness to switch when you feel yourself either waiting needlessly or hitting a wall, is most of the skill. It’s a tiny habit with an outsized payoff — and once it’s automatic, you’ll wonder how you ever worked without it.

If you want the broader picture of how Claude compares to other assistants and tools, see Claude vs ChatGPT vs Gemini and Claude Code vs Cursor vs Copilot.