Effort level: max

For when correctness matters more than cost or speed

This is the top of the range, meant for situations where getting the answer right matters more than what it costs or how long it takes to arrive — genuinely high-stakes decisions, work where a wrong answer is expensive enough that near-unlimited reasoning depth is worth paying for regardless of the premium over every other tier.

Diminishing returns are documented, not hypothetical

Unlike some tiers where the risk is under-provisioning, the specific, real risk at the top of the range is the opposite: reasoning depth here doesn't scale linearly with quality, and overthinking on tasks that didn't need this much depth is a documented failure mode, not a theoretical concern. Applying this tier reflexively to tasks that don't clear the bar for it wastes real spend without a corresponding quality gain, and can occasionally produce a worse answer than a lower tier would have on a task that was simple enough that extensive deliberation actively worked against it.

Test against the tier below before committing

Because the gain over the next tier down isn't guaranteed on every task that looks difficult, the reliable way to decide is a direct comparison — run the same task at both tiers and weigh the actual difference in output against the actual difference in cost, rather than defaulting here because a task feels important.

Related

See effort level: xhigh for the tier most often compared against this one, and the effort & thinking cost estimator for the cost gap between them on your specific task size.