Cloudflare Cuts AI Code Review Inference Cost 68% Through Caching and Model Changes

Cloudflare reduced the per-review inference cost of its AI code reviewer from $2.43 to $0.77 by fixing a prompt cache breakpoint, removing unneeded model calls, and lowering model tiers and effort where quality held. The caching fix alone raised first-request cache hit rates from 14–26 percent to between 91 and 99.5 percent. Roughly 40 percent of the savings came from code changes and 60 percent from model and effort choices. Median review time dropped from about 4 minutes 9 seconds to 1 minute 32 seconds, while acted-on findings per 1,000 reviews rose from 229 to 321. The cost growth had been driven largely by a code quality reviewer that consumed heavy cached context per merge request, and the optimizations were verified with a quality harness.