Most developers never touch the max_tokens parameter, and their Claude bills reflect it.
The default 4096 tokens are often overkill. A small SaaS team in the EU (case-002) was spending $1,840/month on customer support auto-replies—short, repetitive prompts that rarely needed more than 500 tokens. After adjusting, their bill dropped to $287 while maintaining identical quality in blind A/B tests. An indie hacker (case-001) saw similar savings, cutting costs from $312 to $74 on CRM UI codegen by trimming unused token headroom.
Even everyday use adds up. A global developer (case-006) running a Cursor-style assistant daily reduced their $145 bill to $29 by matching max_tokens to actual response lengths. The pattern holds: default settings waste budget on padding, not output.
Audit your own Claude usage. Paste your last 30 days at aiusage.ai—no signup for the number.