What model does Claude Code use?
There is no single model behind Claude Code, and that's by design. It runs the recommended model for your account type — or your organisation's default, if an admin set one — and you override that per session or permanently. Selection normally happens through aliases like sonnet, opus or fable rather than model IDs, so your configuration survives the next release.
- Default
- Account-based or org default
- Switch in session
- /model alias, name, or picker
- Per session
- --model CLI flag at launch
- Permanently
- settings the model field
- Everyday alias
- sonnet daily coding tasks
- Hardest work
- fable not the default
- Long context
- sonnet[1m] 1M-token variants exist
- Fallback
- --fallback-model when the default is unavailable
The docs define the default relatively, and so do we.
Anthropic's model configuration page describes the default alias as the value that "clears any model override and reverts to the recommended model for your account type", or to the organisation default where an admin has set one. That is the honest answer to this question: the default is a policy, not a fixed model name, and it differs between a personal Pro account and a managed Enterprise one.
Selection is through aliases: sonnet for daily coding, opus for complex reasoning, haiku for simple fast tasks, best for the strongest your organisation can reach, plus long-context variants such as sonnet[1m] and opus[1m]. Aliases resolve to whatever the latest model in that family is, which is exactly why they exist.
One named model the docs do pin down: Claude Fable 5 is described as the most capable model in Claude Code, suited to tasks larger than a single sitting — and explicitly not the default. You select it with /model fable. It requires Claude Code v2.1.170 or later, and it is not available under zero data retention, where the picker omits or disables it.
We deliberately don't print "the default model is X" anywhere on this page. Model line-ups move every few months; an answer written that way is wrong within weeks and looks authoritative while it's wrong. Check yours with /model instead.
How to see and change yours.
In rough order of scope:
- In session —
/model <alias|name>switches immediately, or plain/modelopens a picker. If the conversation already has output it asks for confirmation, because the next response re-reads the full history without cached context — that's a real cost, not a formality. - At launch — the
--modelflag, for a single session or a script. - Permanently — the
modelfield in your settings file. - Organisation-wide — an admin default that
defaultreverts to.
This is where a model question becomes a cost question. Rates differ several-fold between families, and an agent reading a repository consumes far more input than a chat does — so running everything on the most capable model is a decision with a number attached.
Current per-token rates for every Claude model are on our Anthropic pricing page, derived from the snapshot rather than retyped here. If you're on a subscription rather than credits, the equivalent lever is your plan's usage allowance — see what Claude Code costs.
You can pass a full model name, and there are cases for pinning one — reproducibility, or a workflow validated against a specific version. For everything else the alias is the better default: sonnet keeps meaning "the current Sonnet" after the next release, while a pinned ID quietly keeps you on last year's model.
There's also --fallback-model, which switches automatically when your chosen model is unavailable — worth setting in CI, where a failed run costs more than a slightly different model.
Picking by cost, not by vibe.
The alias you set decides both quality and price. Two places to check the arithmetic before defaulting to the strongest option available.
Frequently asked.
Quick follow-ups people search after this question.
Q · 01 How do I check which model my session is using? +
/model with no argument — it opens the picker showing what's selected and what's available to your account. The model is also shown above the prompt when a session starts.Q · 02 Can I make Claude Code always use a specific model? +
model field in your settings file, or pass --model at launch for one session. Prefer an alias like sonnet over a pinned model name unless you specifically need reproducibility.Q · 03 Which model should I use for coding? +
sonnet at daily coding, opus at complex reasoning and haiku at simple fast tasks, with Fable 5 for work larger than a single sitting. The honest tiebreak is cost: check the rates before defaulting to the strongest.