Effort level: high

The baseline, not a special case

High effort is what you get by default if you omit the parameter entirely, and it's the recommended floor for anything intelligence-sensitive — not a tier to reach for only when a task looks hard, but the level everything else in the range is measured against. Most tasks that don't have a specific, deliberate reason to run cheaper or push harder should sit here rather than at either end of the range.

Why "no specific risk" doesn't mean "always correct"

Unlike every other tier in the range, this one carries no documented specific risk of its own in the guidance this site sources — it's the baseline, not a tier defined by a known failure mode to watch for. That doesn't mean every response at this tier is automatically correct; it means the known risks that apply to the tiers above and below it (underthinking at the low end, diminishing returns at the high end) aren't the concern here specifically.

The tier to return to after experimenting elsewhere

If a workload has drifted toward either the cheap end of the range in pursuit of cost savings, or the expensive end chasing marginal quality gains, this is the tier worth returning to as the control group — the level to compare any deviation against, since it's what "no deliberate optimisation in either direction" actually looks like.

Related

See effort level: medium for the cost-conscious tier below this one, and effort level: xhigh for where the range goes above it on models that support it.