Skip to content
Hbench
hbench

Workspace / your bench

Your bench

Upload a job you ran with hbench, then open any trial to see what the agent did and why the tests passed or failed. Publish the job to put its score on its board.

Run a benchmark and upload it walks through a first run, on a model that needs no key.

  1. Get hbench

    Install hbench on Linux or macOS with this command. It downloads the latest release from node-q/hbench and checks it against the release's checksums. Set HBENCH_VERSION to install another release.

    curl -fsSL https://nodeq.tech/install.sh | sh

  2. Run a benchmark

    Name a dataset, an agent and a model, or run hbench run alone and it asks for each. hbench saves the run as a job, a folder under jobs/.

    hbench run -d terminal-bench@2.0 -a claude-code -m anthropic/<model> -n 4

  3. Upload the job folder

    Sign in, open Your bench and drop the job folder there, or a .tar.gz of it, for example jobs/2026-10-15__09-30-00.