Every response Claude gives is a tradeoff between speed and depth of reasoning, and the effort level is the setting that decides where on that line Claude works. It sits right next to the model picker, but it does something different: the model decides which Claude you are talking to, and effort decides how hard that model works on your specific request. Get it wrong and either your simple questions burn through more of your usage limit than they need to, or your hard questions come back a little too fast. Here is what each level actually does, and when to reach for it.
What effort levels control
Effort is a dial on how much internal reasoning Claude does before it writes an answer. Anthropic's own support documentation is direct about the cost of turning that dial up: higher effort levels "take longer and use more tokens, so you'll reach your usage limits faster." The dial runs from Low, through Medium and High, up to Extra high and Max, and each step trades speed for a deeper pass over the problem before Claude commits to a response.
This is a separate control from the model picker. Choosing a model decides which Claude family you're using; choosing an effort level decides how hard that model works on this one request. The two settings sit next to each other in the interface, but you can raise or lower either one independently.
The five levels, from routine to maximum
Anthropic documents Low and Medium as the settings for everyday work: they "work well for routine tasks and stretch your usage further." That makes them the right default for short questions, quick edits, or anything where you already know the answer is simple and just want it fast.
High is Anthropic's recommended middle ground. It "offers the best overall balance of quality and speed," and it's a sensible default when you're not sure how hard a task actually is.
Above High sit two levels aimed at genuinely difficult work. Extra high, sometimes labeled "xhigh" in the interface, offers "deeper reasoning than high." Max goes further still, described as giving "the deepest possible reasoning" Claude can apply. Both are worth reaching for on multi-step problems, tricky debugging, or anything where a wrong answer is expensive, but neither is a good default for routine chat.
Effort levels versus model choice
It helps to know that effort and model speed are not the same lever. Anthropic's own model comparison rates baseline response speed by model family, independent of effort level: Claude Fable 5.1 is the slowest, Claude Opus 5 is moderate, Claude Sonnet is fast, and Claude Haiku is the fastest of the family. Effort then sits on top of whichever model you've chosen, adjusting how much reasoning that model does before answering.
One wrinkle worth knowing before you go looking for a toggle: thinking "cannot be turned off in Claude when using Claude Fable 5.1 or Claude Opus 5." Those two models always show their reasoning process, whatever effort level you set. When Claude does think through a problem, you can see the process itself by clicking the "Thinking" section to view Claude's thought process summary, which is useful for sanity-checking an answer on a hard question rather than just trusting the final text.
How to set the effort level
Changing effort takes two clicks: click the model name next to the send button, then click "Effort" to choose a level. There's no separate menu to hunt for, and the setting applies to your next message onward rather than the whole conversation, so you can move between levels mid-conversation as a task gets harder or easier.
If you want to see the effect of a change rather than take it on faith, your usage panel gives a running total: you can check your current usage in Settings, under Usage. Watching that number for a day while you experiment with Low versus High versus Max on similar tasks is the fastest way to build an intuition for what each level actually costs you.
Effort and your usage limits
Usage limits are shaped by more than just how much you chat. Anthropic states plainly that usage is affected by "which Claude model you're chatting with, and the effort level you've selected," alongside the length of your conversations. That means two people asking the same number of questions can burn through very different amounts of usage, purely based on model and effort choices.
The practical takeaway is to treat effort as a budget, not a quality knob you max out by default. Leave routine questions, quick lookups, and simple edits on Low or Medium. Use High as your everyday setting when a task's difficulty is unclear. Save Extra high and Max for the handful of requests each day that actually need deep, multi-step reasoning, where the extra time and usage is worth what you get back.
Conclusion
Claude's effort levels are a straightforward idea with a real cost attached: more effort means more reasoning, more time, and more of your usage limit spent per message. Treating High as your default, dropping to Low or Medium for routine work, and reserving Extra high and Max for problems that genuinely need them is the difference between usage limits that last all day and ones that run out by lunchtime.
Relevant Training
Claude effort levels, alongside model choice, the Thinking toggle, and the tradeoff against usage limits, are covered in full in Starting with Claude, the first course in Cybrec's Claude AI training program.
0 Comments