Two settings shape every answer Claude Code gives you. The model decides which Claude does the work. Effort decides how hard it works on each step: how much it thinks, tests and double-checks before replying. Both trade the same things against each other: speed, tokens and depth.
Choosing a model
Run /model to open a picker, or /model opus to switch straight away; that also saves it as your default for new sessions (press s in the picker to switch for this session only). You can also launch with claude --model opus, set ANTHROPIC_MODEL, or put "model" in a settings file.
Most people use aliases rather than version numbers: opus for complex reasoning, sonnet for daily coding, haiku for simple tasks, and fable for the hardest, longest-running work. Aliases move to newer versions over time; to pin one, use a full name such as claude-opus-5-5. Two special values: default clears your override and uses your account’s default model, and opusplan uses Opus while you’re in plan mode and Sonnet once it starts executing.
A model switch also reaches subagents that inherit the session’s model, unless their definition sets its own.
Choosing effort
On models that support it, effort controls adaptive reasoning: the model decides on each step whether and how much to think. The levels are low, medium, high, xhigh and max, though which ones exist depends on the model. Opus 5.5 and Sonnet 5.5 default to medium; most other models default to high.
Set it with /effort (a slider, or /effort high), the --effort flag, the CLAUDE_CODE_EFFORT_LEVEL variable, or an effort field in a skill or subagent’s frontmatter. For a single hard turn, put ultrathink in your prompt instead of changing the level. Unless you set it through that environment variable, max lasts for the current session only.
Need it faster rather than deeper? /fast turns on fast mode: the same Opus model, configured for faster output at a higher price per token.
The catch
Higher effort means more thinking tokens, and you pay for those even when they’re hidden. The docs warn that max can overthink. The same level name also isn’t calibrated the same way on every model, so retest when you switch rather than carrying a level over. A practical split: low effort while you’re steering each step, higher when Claude must verify on its own.