What Claude Code's multi-agent code review costs and finds

Quickchat's benchmark found that Claude Code's multi-agent review matched slightly more serious issues than a single agent, at a much higher recorded cost.

Across 20 pull requests on Sonnet 5.5, max matched 23 of 27 serious answer-key issues, versus 21 for high. Finished max reviews averaged 56 minutes and $20.95, versus 40 seconds and $0.11 for high. Both matched 47 of 65 total answer-key issues. Quickchat says the serious-issue gap is about the scoring noise.

The test used Claude Code 2.1.303's local /code-review command, not the managed service. Max used custom limits: Sonnet 5.5 for every subagent, $30 per review and 90 minutes. Only 19 of 20 max reviews finished. The sample was selected from PRs where Quickchat's own reviewer had performed worst.