cognirank

cognirank

Ranks small AI models on reasoning and programming benchmarks, with sortable tables, charts and side-by-side comparisons, as a static website. For people choosing or studying small models.

Made by dotlabs on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: cognirank is a static website that ranks small AI models on reasoning and programming benchmarks. It offers rankings you can sort, charts and model comparisons, plus a news feed you edit in a JSON file. It reads a live registry first and falls back to a bundled snapshot.

Who it suits: It suits people who need to compare small models and anyone who wants a leaderboard site they can read and adapt.

Industry | Who uses it | How they use it | Example
AI research | Researchers | Compare small models on reasoning and coding benchmarks | Sort the rankings by a score you care about
Software | Developers | Shortlist a small model for an app | Put two candidate models side by side
Education | Students | See how small models stack up | Browse the charts
Web publishing | Site maintainers | Run a static leaderboard with a news feed | Add a news slide by editing news.json

Skip it if you need to run the benchmarks yourself, since the site shows results and does not evaluate models.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• A local copy served from your own computer on Windows, macOS or Linux
• Any static web host, with the whole folder uploaded and its paths kept

Use case | Who | What they get back
Compare small models | Researchers | Sortable rankings and charts from the evaluated snapshot
Keep the site current | Maintainers | News slides from entries you add to news.json
Host your own copy | Developers | A site made only of HTML, CSS, JavaScript and JSON files

How to set it up: The online page needs nothing. A local copy needs Python, both to fetch the files and to serve them over HTTP.

• Python and pip
• Windows, macOS or Linux
• A web browser

1. Install the Hugging Face hub library.
2. Download the Space's files into a folder named cognirank.
3. Move into that folder.
4. Serve the folder over HTTP, because pages opened straight from disk generally cannot fetch local JSON files.
5. Open port 8000 on your machine in a web browser.

pip install -U huggingface_hub
hf download dotlabs/cognirank --repo-type space --local-dir cognirank
cd cognirank
python -m http.server 8000

How to use it: Browse the rankings, sort them and compare models. For news, put a new object into the news array of news.json, placing the latest item at the top. Changes show up the next time the page loads, with no HTML or JavaScript edits.

Field | What it holds
id | A short name for the slide
date | The day of the update, written as year, month, day
title | Your announcement headline
description | A short update
image_url | The slide image, as a web address or a local path
image_alt | Text that describes the image

Tool | What it does
index.html | The page markup and nothing else
css/styles.css | All styles and the responsive layouts
js/app.js | Rankings, charts, sorting, comparisons and the JSON-driven updates
data.json | The evaluated snapshot, used when the live registry cannot be reached
news.json | The news slides you edit

Example: You add a new entry to news.json with today's date and a short title, then reload the page. The news slide appears without any change to the HTML or JavaScript.

For newcomers and experts: Open the page and sort the rankings by one column to see how small models differ. Nothing needs installing.

The page asks the live dotlabs/cognirank registry first and uses the bundled data.json if that request fails. To make the bundle primary, swap the two entries in the source list inside loadModels(). The mobile layout once used virtualization with spacer rows, which could squeeze the Model column to almost nothing on narrow screens. It now uses normal table rows and native scrolling, which is simpler and reliable at this size.

The page loads
It asks the live registry for model data
If that fails, it reads data.json
Rankings and charts are drawn
News slides come from news.json

Good to know:

• Hosting: Publish the entire folder together and keep its paths.
• Limit: Opening index.html straight from disk may leave the data blank, so serve the folder over HTTP.
• License: cc-by-nc-nd-4, with a LICENSE file in the repository. Check its terms before you reuse or change the site.
• Cost to run: none named.
• Data it reaches: The live registry on Hugging Face first, then the bundled data.json. News images may come from outside addresses.

On Hugging Face, 13 November 2026

• Likes: 15
• Built with: static
• License: cc-by-nc-nd-4.0

Links

• Space: https://huggingface.co/spaces/dotlabs/cognirank
• Files: https://huggingface.co/spaces/dotlabs/cognirank/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.