Claude Code models: which ones you can use and how to switch

Which models can Claude Code use?

Claude Code picks a model through an alias, so you do not have to remember version numbers. The alias points at the newest recommended version for your provider and moves forward when Anthropic ships a new one. This table is from the model configuration page of the Claude Code docs, checked in October 2026.

Fable is never the account default. You choose it yourself. The docs describe Fable 5.1 and Fable 5 as the most capable models in Claude Code, suited to tasks larger than a single sitting.

Other providers resolve the aliases to older models. On Amazon Bedrock and Google Cloud's Agent Platform, opus is Opus 5.5 but sonnet is Sonnet 4.5 and haiku is Haiku 4.5. On Microsoft Foundry the defaults are Opus 4.6, Sonnet 4.5 and Haiku 4.5, and the default model there is Sonnet 4.5. If you want the newer model on those providers, select it by its full name or set ANTHROPIC_DEFAULT_OPUS_MODEL or ANTHROPIC_DEFAULT_SONNET_MODEL.

How do you switch models in Claude Code?

There are five ways to set the model, listed here from highest to lowest priority. The first one that applies wins.

1. During a session: type /model followed by an alias or name, such as /model sonnet. Run /model with no argument to open the picker.
2. At launch: start with claude --model opus (or any alias or full name).
3. Environment variable: set ANTHROPIC_MODEL to an alias or name in your shell.
4. Settings file: put a model field in settings.json. This is the permanent option.
5. Default for new sessions: set ANTHROPIC_DEFAULT_MODEL. It applies only when none of the four above selects a model, and it needs Claude Code v2.1.236 or later.

The picker has one detail worth knowing. Pressing Enter on a row switches the model and saves it as your default. Pressing s switches for the current session only and leaves your saved default alone. Typing /model sonnet directly behaves like Enter, so it saves.

# Start a session on Opus
claude --model opus

# Switch inside the session (also saves it as your default)
/model sonnet

# Open the picker; press s on a row to switch for this session only
/model

# Pin an exact version instead of an alias
claude --model claude-opus-5-5

# Fable 5.1 through its alias
/model fable

# Check which model you are on
/status

Three behaviors catch people out. First, a session you resume with claude --resume or --continue usually keeps the model it was using, not your current default. Second, if you switch with /model, subagents that inherit the main conversation's model switch too, because their model is resolved from the one your session uses when Claude starts them. To keep a custom subagent on a smaller model, set the model field in its definition (see the subagents guide). Third, to run different models in different terminals at the same time, launch each with its own --model flag instead of switching with /model.

How do you set a default model in settings.json?

Add a model field to a settings file. The user file at ~/.claude/settings.json applies to every project on your machine. The project file at .claude/settings.json applies to everyone who works in that folder, and the local file .claude/settings.local.json is yours alone for one project. The permissions guide explains how these files stack.

{
  "model": "opus",
  "permissions": {
    "allow": ["Bash(npm run lint)"]
  }
}

The docs give this exact shape, with a model next to a permission rule. A value can be an alias or a full model ID. The /model command also writes this field into ~/.claude/settings.json when you save a choice.

If a new session starts on a different model than the one you picked, the docs list the usual causes. You chose it for one session only (the s key, --model, or /model in non-interactive mode). Something with higher priority sets the model, such as a model value in project or managed settings, or ANTHROPIC_MODEL in your shell. The file ~/.claude/settings.json is not writable. Or you resumed a session that kept its old model. When project or managed settings set the model, the startup header names the file.

Which model should you use for which job?

The docs' own guidance is short. The default, Opus 5.5, handles most coding well. Sonnet costs less on routine work. Fable is there when you select it explicitly. Haiku is a good fit for subagents doing simple jobs. The table turns that into choices for a builder working through an app idea.

Two cost notes from the docs. Depending on your plan and seat tier, Fable usage can bill to usage credits instead of your plan's included limits. When it does, the /model picker shows “Requires usage credits” on the Fable row. And requests that a Fable model's safety classifiers flag, most often in cybersecurity and biology topics, trigger automatic model fallback. For plan details see the Claude Code pricing guide and the usage limits guide.

If you are starting a project from a build prompt, a sensible split is to run the first planning pass in plan mode on opusplan, review the plan, then let the session build. The guide to building an app with Claude Code walks through that flow, and every idea on the board comes with a ready prompt to paste.

What is effort, and how is it different from the model?

The model decides who does the thinking. Effort decides how much thinking happens. Effort controls adaptive reasoning: the model chooses whether and how long to think on each step, and a lower level is faster and cheaper while a higher level reasons more deeply. The docs say effort has replaced most of what people used to do with prompt wording.

Set it with /effort and a level name, or run /effort alone for a slider. Use /effort auto to clear your saved level for the active model. In the picker, the left and right arrows adjust effort. Enter saves the level as your default and s applies it to the current session only. The max level is session-only unless you set it through the CLAUDE_CODE_EFFORT_LEVEL variable.

For one hard question in the middle of routine work, put the word ultrathink in that prompt instead of changing the setting. The docs say other phrases such as “think hard” are ordinary prompt text, not keywords. Opus 5.5, Sonnet 5.5, Haiku 5.5 and the Fable models also cannot have thinking switched off.

How big is the context window for each model?

Fable 5.1, Fable 5, Sonnet 5 and later, Haiku 5.5, and Opus 4.7 and later run with a 1 million token window by default, with no [1m] suffix needed. Opus 4.6 and Sonnet 4.6 reach 1 million only through their [1m] variants, and access depends on your plan. On the Anthropic API the 1 million window uses standard model pricing with no premium for tokens beyond 200K, except Haiku 5.5, which costs more on prompts longer than 100K tokens.

To keep sessions at a 200K window, set CLAUDE_CODE_DISABLE_1M_CONTEXT=1. A bigger window does not mean you can stop managing context. A long conversation is re-sent with each request, so it still costs more. The context window guide covers compaction and what to do about it.

How do you pin a version or set a fallback model?

Aliases move. If you need a repeatable setup, such as a team project or a CI job, pin the full name, for example claude-opus-5-5, or set the matching environment variable: ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL or ANTHROPIC_DEFAULT_HAIKU_MODEL. A mistyped value set through --model, ANTHROPIC_MODEL or the model setting is not checked at startup. You see an error on the first request.

When the primary model is overloaded or returns a non-retryable server error, Claude Code can switch to a fallback model for that turn. Pass --fallback-model with a comma-separated list, or set fallbackModel as an array in settings. Chains are capped at three models, and the switch lasts for the current turn only.

claude --model opus --fallback-model sonnet,haiku

Administrators can restrict which models appear with availableModels and block specific ones with deniedModels in managed settings. If your picker is missing a model you expect, an allowlist is the first thing to check.