Skip to main content
Claude Code supports multiple Claude models and lets you switch between them interactively or via configuration. Each model has different speed, capability, and cost tradeoffs. Effort levels further tune how much thinking a model applies to a given task.

​
Selecting a model

​
Interactive model picker

Run /model with no arguments to open the interactive model selection menu. Use arrow keys to pick a model and, for supported models, adjust the effort level. 
/model

​
Set model directly

Pass the model name or alias as an argument: 
/model sonnet
/model claude-sonnet-4-6
/model opus
/model default
Passing default resets the session to the configured default model.

​
Persist a model in settings

To use a specific model in every session, set model in your user or project settings:
~/.claude/settings.json
{
  "model": "claude-sonnet-4-6"
}
See Settings for all settings file locations and precedence rules.

​
Model aliases

Claude Code recognizes short aliases that always resolve to the latest recommended version of that model family:
AliasResolves to
sonnetLatest Claude Sonnet
opusLatest Claude Opus
haikuLatest Claude Haiku
bestBest available model
sonnet[1m]Claude Sonnet with 1M context
opus[1m]Claude Opus with 1M context
opusplanOpus model used in plan mode
Aliases are defined in utils/model/aliases.ts and are resolved at session start. Using an alias means your configuration follows model updates automatically.
Extended context ([1m]) models require account access. Claude Code will show an error if your account does not have access when you attempt to select one.

​
Effort levels

Effort controls how deeply the model reasons before responding. Not all models support the effort parameter — it is currently available on claude-opus-4-6 and claude-sonnet-4-6.

​
Available levels

low

Quick, straightforward implementation with minimal overhead. Best for simple edits and fast iteration.

medium

Balanced approach with standard implementation and testing. The recommended default for most tasks.

high

Comprehensive implementation with extensive testing and documentation. Used when thoroughness matters.

max

Maximum capability with deepest reasoning. Opus 4.6 only. Session-scoped — not persisted to settings for external users.

​
Set effort via command

Use /effort to change the effort level for the current session: 
/effort low
/effort medium
/effort high

​
Persist effort in settings

Persist an effort level in settings.json using the effortLevel key: 
{
  "effortLevel": "medium"
}
"max" effort is only supported on Opus 4.6. On other models it is automatically downgraded to "high". For external users, "max" is session-scoped and never saved to settings.json.

​
Effort precedence

The effort value applied to each API call follows this precedence chain: 
CLAUDE_CODE_EFFORT_LEVEL (env var)  →  session effortValue  →  model default
Setting CLAUDE_CODE_EFFORT_LEVEL=unset in your environment disables the effort parameter entirely, letting the API apply its own default.

​
Default effort by model

Default effort levels are determined at runtime and vary by subscription type:
  • Opus 4.6 + Pro subscription: defaults to medium
  • Opus 4.6 + Max/Team (when configured): defaults to medium
  • Other supported models: no default (API applies high internally)

​
Enterprise model controls

Administrators can restrict which models are available to users via the availableModels setting in managed settings:
managed-settings.json
{
  "availableModels": ["sonnet", "claude-sonnet-4-6"]
}
If a user attempts to select a model not in the allowlist, Claude Code will display an error and keep the current model. An empty availableModels array restricts users to only the default model. Model IDs can also be remapped to provider-specific identifiers (e.g. Amazon Bedrock ARNs) via modelOverrides:
managed-settings.json
{
  "modelOverrides": {
    "claude-opus-4-6": "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-opus-4-6-v1"
  }
}

Build docs developers (and LLMs) love

Get started for freeTalk to us