Cut your Claude Code bill without losing quality: what 100 measured runs say about effort levels

Description and use

rightsize cuts Claude Code costs up to 98 % without losing quality by applying rules for model and effort selection.

Problem

Claude Code’s weekly limits were cut, causing users to pay for unnecessary token usage while quality does not improve beyond a certain effort level.

Outcome

rightsize reduced cost by 73‑98 % on three benchmark tasks while keeping the same pass rates, and showed that effort above medium adds no measurable quality on verifiable tasks.

Workflow

• Score each task 0–3 on scope, risk, ambiguity and depth.
• Select model and effort based on the score (Haiku/low for lookups, Sonnet/medium for standard code, Opus for risky changes).
• If a check exists, start one level lower and only increase effort if the check fails twice or a high‑severity review occurs.
• Keep high effort for work without a check (design, review, framing).
• Test the change’s reach during work and run the full suite at sprint end, logging full runs.
• Protect the cache by not switching the main model mid‑session and routing work through sub‑agents.

Evidence

Since mid‑September 2026, Claude Code's weekly limits have been cut and reports of quotas drained in a day keep piling up.

Source

rightsize turns that into rules: Score every task 0–3 on scope, risk, ambiguity and depth.

Source

Above medium, effort bought nothing measurable on these tasks. It only raised cost and time.

Source

Opus at high everywhere vs rightsize on three tasks: same pass rate, cost cut by 73% to 98%.

Source