Show HN: Benzi – A Code Intillegence/Harness Beating Claude Code and CodeGraph

Category: devtools

Tags: code-intelligence, benchmark, ai-coding, swe-bench, harness

Score: 6.0/10 (Innovation: 7, Technical: 6, Documentation: 4, Utility: 5)

Benzi is a code intelligence harness that claims to outperform Claude Code and CodeGraph on SWE-bench Verified, resolving 82.4% of issues at 3.1¢ per fix. It emphasizes apples-to-apples benchmarking with transparent metrics like lines read, wall clock, and cost. The project is interesting for its novel evaluation methodology and potential cost-efficiency, but lacks detailed technical documentation and open-source details.

Target audience: AI researchers, backend developers, devops, engineering managers

Repository: https://benzi.fly.dev/benchmark

View on Hacker News