LocalArena

Open source · Apache-2.0

Your models.
Your tasks.
Your evidence.

Run live comparisons across local or hosted models. Keep every result portable, private by default, and honest about uncertainty.

Star on GitHub Read the docs
Quick start Python 3.10+
01 Install
pip install localarena
02 Run a live score
localarena quickstart ollama qwen3:0.6b

Why LocalArena

A decision you can reproduce.

  1. 01

    Live, not recorded

    Call the exact local or hosted endpoints you choose. No copied leaderboard answers and no silent fallback.

  2. 02

    Your workload stays yours

    Evaluate private task packs on your own machine. Prompt and answer content is excluded from saved results by default.

  3. 03

    Evidence before rankings

    Preserve failures, coverage, latency, and uncertainty so a narrow win is never presented as certainty.

Bring any compatible endpoint

Start with one command

Make your own model decision.

pip install localarena Get started