Skip to main content
Benchspan is a platform for benchmarking AI agents. You write a single runner.sh that tells us how to run your agent. We handle everything else — Docker containers, benchmark infrastructure, scoring, and analytics.

​
How it works

Your agent  →  runner.sh  →  Benchspan  →  Results
  1. You write a runner.sh that installs and runs your agent
  2. We package it and run it against benchmark tasks in isolated Docker containers
  3. Each task gives your agent a problem statement and a working directory
  4. Your agent solves the problem, we grade the result
  5. You get scores, token usage, latency, and detailed logs

​
Get started

Quickstart

Install the CLI, log in, and run your first benchmark in 5 minutes.

Onboard your agent

Write a runner.sh for your agent and verify it works with the healthcheck benchmark.

Built-in agents

Run pre-configured agents like Claude Code or OpenHands without writing any code.

CLI Reference

Full reference for all Benchspan CLI commands.