Leave Effort at high

For most development work, use high. It is already the default on every current model (Fable 5.1, Opus 5, Sonnet 5), so you can usually leave it alone.

There is rarely a reason to go higher unless the task is unusually difficult and Claude has already struggled.

Effort	When to use
low / medium	Simple tasks where speed matters most.
high	Default for nearly all development work. Balances token usage and intelligence.
xhigh	Deeper reasoning at higher token spend, for genuinely hard problems.
max	Rare exception. It can show diminishing returns and is prone to overthinking, so test before relying on it.

Available levels depend on the model. The 4.6 models have no xhigh.

Setting it

Use /effort <level> in a session. To make it durable, set effortLevel in settings to low, medium, high, or xhigh. max is not accepted there: it applies to the current session only, unless you set it through the CLAUDE_CODE_EFFORT_LEVEL environment variable.

The /effort menu also offers ultracode. That is a Claude Code setting, not a model effort level: it sends xhigh and has Claude orchestrate multi-agent workflows.

Practical advice

1. Start with high.
2. Review the result.
3. Increase effort only if the problem is genuinely hard or the first attempt exposed reasoning gaps.

On most models, changing effort mid-session re-reads the whole conversation uncached, so raise it in a fresh session when you can. See Prompt cache pitfalls.