kodo bench — join

Join the benchmark

Help evaluate AI coding agents on SWE-bench. Run benchmarks on your machine, results upload automatically to the shared leaderboard.

Display name

Benchmark guidelines

Run agents with the standard benchmark harness as-is
Do not modify the evaluation pipeline or test outcomes
Do not fabricate, cherry-pick, or alter results
Each agent must work autonomously — no human-in-the-loop
All submissions are public and attributable to you

I agree to follow these guidelines and understand that my submissions are public and tied to my identity.

You're in!

Save this token — it cannot be retrieved again.

click to copy

1. Install

uv tool install --with 'kodo[benchmark]' git+https://github.com/ikamensh/kodo

2. Configure

3. Run

kodo-bench

The harness auto-detects which agents you have installed (Claude Code, Cursor, Codex, Gemini) and requests task assignments from the server. Results upload after each task.

To run specific agents: --backends claude,cursor