HN Explorer

Show HN: Try Benzi – A coding harness/agent beating Claude Code itself on Sonnet (benzi.fly.dev)

8 points by showhz ↗ HN
Hi y'all. Been working on something that should've been made a long time ago imo. It compiles codebases into O(1) hashmaps that the agent queries to discover the structure of your code/answer questions/write code.

It also does complete static analysis checks on any writes the agent makes.

Don't take my word for it though. Here are the benchmarks: https://benzi.fly.dev/benchmark. on 2/20 tests, Claude Code (mostly Sonnet on one task) regressed or timed out. Benzi didn't because of course, it has a map it can query and not get lost in the sauce. On the other 18 it is cheaper, faster, or often both.

Would love to get some early adoption and criticism!

(Only available on Windows for now. soz. and also keep an eye on the benchmarking page; I think I can push it far more -- no promises, still work in progress. Haven't thoroughly tested greenfielding experience either.)

(another note: VS Code extension/website is running DeepSeek V4 flash. not Sonnet. Everything mostly built with CC Sonnet tho )

4 comments

[ 4.2 ms ] story [ 24.4 ms ] thread
showhz
hi. rewrote the text for more visibility. Benzi remains the same product:

Benzi is a code intelligence software.

It starts with a tree sitter, and produces a unique Benzi ID for all symbols in arbitrarily large codebases.

With all callflow and dataflow resolved, the agent simply queries the codebase using agentic tools instead of reading though it using traditional RAG approaches or trying to "rank" results using embedding-space approaches.

This affords faster inference, cheaoer prices, and less context drift. Tested over 24 repos, Benzi reads 2x less source code than Claude Code, 3x less than the very recently released DeepSeek Harness, and 6.5x less than OpenCode.

Benzi has several bonus features such as syntax+semantic verified code edits, runtime tracer to track actual execution through the Benzi compiler, context aware model generated repros to fast track testing, etc.

On the benchmarks side, Benzi + DeepSeek V4 Flash scores 78% on SWE-bench Verified (benchmark details on the benchmark page). For comparison, DeepSeek reports 73.7% as the baseline scaffolding number for v4flash and self reports their own score to be 78.6%. HOWEVER. Benzi gets to this parity reading 3x less source code than the deepseek harness. (benchmarks page again)

Please try it out, and let me know what you think!

https://benzi.fly.dev/

https://lnkd.in/gtiv7feC