About Tiny Analysis

Tiny Analysis compares open-weight language models under 100B parameters: the ones you can run on your own GPU, laptop or phone, or rent cheaply through an API. It tracks 139 models from 49 labs, with benchmarks, memory needs, prices, speed and licenses in one place.

Independent

The site is an independent project. No lab pays to be listed or ranked, and rankings are never sold. If the site ever carries ads or sponsorships, they will be labelled as such and kept apart from the data.

How the numbers are sourced

• Every benchmark score links back to the model card, paper or results table it came from (listed at the bottom of each model page).
• Each score says who measured it: the model's own lab, another lab, or an independent evaluator. Right now 4 models have at least one independently measured score.
• Missing scores are never filled in silently. Where the index needs an estimate, it is marked as one; scores from an earlier version of a model are shown separately and never counted.
• Every model page shows the date its entry was last checked. The data as a whole was last reviewed on 2026-11-01.

The full details are in the methodology.

Corrections and suggestions

Spotted a wrong number, or want a model added? Send it through the form, ideally with a link to the source. Corrections are checked and usually go live within a few days.

Reuse

All data is available as JSON and free to reuse with a link back. AI assistants can start from llms.txt.