Skip to content

models.md

Latest commit

 

History

History
80 lines (58 loc) · 7.44 KB

models.md

File metadata and controls

80 lines (58 loc) · 7.44 KB
Raw

Models and effort

Verified against Claude Code v2.1.309 (stable) on 2026-11-02

Which model a session runs and how hard it thinks are two separate choices. This page is the lookup for both: the model aliases, the effort levels and the ways to set them. Which model and effort a task needs is taught in the Intermediate lesson Pick the model and effort. To compare the current models and find their IDs, see Anthropic's models overview.

Aliases

An alias picks the recommended model of a family, so you never need a version number. Pick one with /model <alias>, claude --model <alias> or the model setting (Model configuration).

Alias On the Anthropic API it uses Reach for it when
sonnet The latest Sonnet Day-to-day coding
opus The latest Opus Complex reasoning
fable The Fable model for your provider The hardest, longest-running tasks
best Fable where your account has it, otherwise Opus You want the strongest model available to you
haiku The latest Haiku Quick, simple tasks
opusplan Opus in plan mode, then Sonnet to carry out the plan You want deep planning and cheaper execution
opus[1m] Opus with a 1 million token context window Very long sessions, on a provider where opus doesn't already have that window
  • An alias moves to a newer model over time. To stay on one version, use its full model name, such as claude-opus-5-5, from the models overview.
  • default isn't an alias: it clears any model you picked and returns to your account's default.
  • On other providers an alias can resolve to an older model: sonnet and haiku do on Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry, and opus does on Microsoft Foundry. The alias table lists them.
  • Fable can draw on usage credits depending on your plan and seat (Fable and usage credits).

The default model differs by account type and provider. Check yours with /model or /status, and see Default model setting for the rules.

Effort levels

Effort controls how much the model reasons on each step. Lower effort answers sooner and spends fewer tokens; higher effort checks more of its own work (Adjust effort level).

Level Fits
low Quick exchanges where you review each result: a first sketch, a rename
medium Day-to-day work with a clear scope, such as a new feature, on Opus 5.5 and Sonnet 5.5; on models with a higher default, a cheaper level that gives up some capability
high Work where edge cases are likely, such as fixing a bug in existing code
xhigh Deeper reasoning at a higher token spend
max Hard problems Claude works through without you; test it before relying on it, since it can overthink
  • Which levels a model offers, and which one it starts at, differ by model: see the levels per model in the docs. /effort status prints the level in effect.
  • Set a level the model doesn't offer and Claude Code uses the highest level it does offer below that.
  • The scale is calibrated per model, so high on one model is not the same amount of reasoning as high on another.

Setting effort

Where How long it lasts
/effort <level>, or /effort for a slider Saved as your default for that model; press s in the slider for this session only
The effort slider in /model The same as /effort
claude --effort <level> This session
CLAUDE_CODE_EFFORT_LEVEL environment variable Every session started with it. It overrides the other sources, though a maxEffortLevel cap still applies
modelSettings or effortLevel in a settings file Every session the file applies to. modelSettings sets one model's level; a top-level effortLevel in your user settings doesn't apply to the newest models
effort in a skill's or subagent's frontmatter While that skill or subagent runs

/effort auto clears the level you saved for the current model; the model then uses an effortLevel from your settings if one applies, otherwise its own default (Set the effort level).

Gotchas

  • max applies to the current session only, unless you set it with CLAUDE_CODE_EFFORT_LEVEL. The effortLevel and modelSettings settings don't accept it.
  • ultrathink in a prompt asks for deeper reasoning on that turn only and leaves the effort level unchanged. Phrases such as "think hard" are ordinary text (Use ultrathink).
  • Ultracode is a setting, not an effort level: with it on, Claude plans a dynamic workflow for each substantive task, at whatever level the session runs (Model configuration).
  • Switching models in the middle of a session recomputes the whole request, and on most models so does changing effort, so Claude Code may ask you to confirm first (Prompt caching).

Related pages

The guide carries no prices or plan limits. See claude.com/pricing for plans and API pricing for per-token rates.

Sources: Model configuration · Models overview · Prompt caching · Fast mode · Model deprecations · Pricing · API pricing

Up: Reference