OpenWeights Bench

Published by Vincony AI Ltd.

openweightsbench.com Visit OpenWeights Bench

What it covers

Benchmark results for open-weight AI models: each model is run locally on one laptop through a fixed battery of 270 items in 13 categories, with scores, speed and memory profiles and written reviews by two judge models.

Who it is for

People comparing open-weight AI models to run on their own hardware.

Countries

The site does not say which countries it covers.

Where its information comes from

  • The site's own test runs
  • Download rankings from a public model-hosting hub, used to choose which models to test
  • Two paid AI judge models, named on the site

How often it is updated

The site says new models and new versions of existing models are tested as they appear.

Cost

Free

On the pages checked, the site states no price, advertising or sponsorship. It says its results and scores may be reused, including commercially, with credit.

Known limits

  • "It is not a leaderboard of overall model quality."
  • "The battery is small, runs at one quantisation and one seed, and judge models are fallible."

Corrections

The site's own route: corrections page on openweightsbench.com.

Email: hello@openweightsbench.com

Questions or corrections:

You can also report an error to the publisher.

Checked on 19 November 2026.