Claude isn’t one thing. It’s a family of models — different sizes, different capabilities, different costs — and Claude Code ships with all of them one click of the model picker away. Using the same model for everything is like using a sledgehammer to hang a picture frame: works, but not the point.
The model is the brain. Bigger means deeper reasoning — but also slower and pricier. Smaller means faster and cheaper — but shallower.
The four models
| Model | Speed | Best for |
|---|---|---|
| Haiku 4.5 | Fastest | Quick edits, lookups, simple one-liners |
| Sonnet 5 | Balanced | Most coding tasks, writing, explaining — the default |
| Opus 4.8 | Deep | Complex architecture, subtle bugs, long multi-file reasoning |
| Fable 5 | Deepest | The hardest problems — state-of-the-art on nearly every benchmark |
Haiku 4.5 is the fast-food order: cheap, instant, good enough for tasks where you mostly know the answer and just want Claude to execute it. Autocomplete-style edits, quick file lookups, “rename this variable across the file” — that’s Haiku territory.
Sonnet 5 is the default for a reason — and the newest version of the Sonnet tier, closing much of the gap to Opus-level reasoning while staying fast and affordable. It handles the overwhelming majority of real coding tasks well — explaining an unfamiliar codebase, writing a new function, fixing a bug you’ve already diagnosed. If you don’t think about which model to use, Sonnet is usually the right call.
Opus 4.8 is the specialist you call in when Sonnet isn’t cracking it. Complex architecture decisions with a lot of moving parts, subtle bugs that have stumped you for an hour, tasks that require holding a large codebase in mind and reasoning across it — these are Opus problems.
Fable 5 is the newest and most capable model Anthropic has made publicly available (released June 2026). It’s built on the same underlying research as Claude Mythos — Anthropic’s frontier-safety work — and is state-of-the-art across software engineering, knowledge work, and scientific reasoning. Reach for it when Opus still isn’t enough: the very hardest design problems, reasoning chains that span an enormous codebase, or tasks where correctness matters more than cost.
A note on Mythos
You may hear about Claude Mythos — the underlying model that Fable 5 is based on, with certain safety restrictions lifted. Mythos is not publicly available; it’s deployed through a partnership program (Project Glasswing) for a small group of cybersecurity defenders and infrastructure providers. For everyday Claude Code work, Fable 5 is the top of the ladder.
How to switch
In the Desktop app there’s a model picker right by the chat box — click it and choose Haiku, Sonnet, Opus, or Fable from the menu. That’s the whole move; no command needed.
Prefer to type? /model does the same thing from the chat. Hit enter and Claude Code pops up the same menu — click the one you want, or start typing the name:
/model claude-haiku-4-5
/model claude-sonnet-5
/model claude-opus-4-8
/model claude-fable-5Either way the switch is immediate. You stay in the same conversation; the next message goes to the new model.
The real skill: knowing when to go up or down
Most people never switch models. They use whatever the default is — and either overspend on simple tasks or get stuck on hard ones because they’re afraid to reach for the bigger tool.
The heuristic that works:
- Stuck after two tries? Go up one tier. If Sonnet can’t crack it, try Opus. If Opus still can’t, try Fable 5.
- Churning through quick tasks? Go down to Haiku. Sonnet on a task Haiku could handle is waste.
- Not sure? Stay on Sonnet. The default is the default because it hits the right spot most of the time.
The signal to go up is “this is harder than I thought.” The signal to go down is “I know exactly what I want — I just need it fast.”
What’s next
You’ve personalized Claude’s knowledge, commands, permissions, and now the brain it uses. In the next level you’ll give it company — spinning up multiple Claudes to work in parallel, delegate subtasks, and take on jobs too big for one conversation.