Hacker News
Show HN: Benzi – A Code Intillegence/Harness Beating Claude Code and CodeGraph
Benzi’s harness outperforms mainstream harnesses on a 24-issue, 10-language SWE-bench set, resolving 78.2% of 500 real bugs for under $0.10 each. Across identical bugs, Benzi reads fewer lines and spends less wall-clock time than Claude Code, whose cost and line count rise steeply with difficulty. DeepSeek runs are an order of magnitude cheaper, but Benzi still achieves the lowest per-bug cost and line-read slope.