HAKARI-Bench Leaderboard

HAKARI-Bench Leaderboard

Shows the public results of HAKARI-Bench, a light multilingual information retrieval benchmark, with tables, model comparisons and filters. For researchers and engineers choosing retrieval models.

Made by hakari-bench 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: HAKARI-Bench is a retrieval benchmark in the style of the Nano sets, and this Space shows its public rankings. It scores BM25 alongside reranker, dense, sparse and late-interaction systems that are compatible with SentenceTransformers. The app loads the latest results database when it starts and serves interactive tables, model comparisons, task notes and filters from it.

Who it suits: Researchers who compare retrieval models, and engineers who need to choose a retriever or reranker for search over text in several languages.

Industry | Who uses it | How they use it | Example
Research | IR researchers | Compare dense, sparse, reranker and late-interaction systems | Filter the tables to one system type
Software | Search engineers | Pick a retriever or reranker for a multilingual product | Look at where bge-m3 sits among the others
Academia | Benchmark authors | See how a new system compares with the listed ones | Check the task notes before running their own
Machine learning | Model builders | Read how models score on each task | Use the filters to narrow the view

Skip it if you only want a single number, since the point of the page is detailed tables and filters.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Docker on Windows, macOS or Linux, for a local copy of the viewer

Use case | Who | What they get back
Compare retrieval models | Researchers | Interactive tables and model comparisons
Read the task notes | Benchmark users | Documentation for each task in the suite
Filter the results | Engineers | A narrowed view of the systems you care about

How to set it up:

• Nothing at all to use the online page
• Python and pip, to download the files with the Hugging Face tools
• Docker, to build and run the viewer
• An internet connection, because the app fetches its results database when it starts

1. Open the Space's page at Hugging Face to use it online.
2. For a local copy, install the Hugging Face hub package and download the Space.
3. Move into the new folder and build the Docker image.
4. Run the container with port 7860 published, then open that port in your browser.

pip install -U huggingface_hub
hf download hakari-bench/leaderboard --repo-type space --local-dir leaderboard
cd leaderboard
docker build -t leaderboard .
docker run -it -p 7860:7860 leaderboard

The repository is tiny, with a Dockerfile, a README and models.py. The viewer itself comes from a ready-made Docker image, which the project's GitHub repository releases.

How to use it: Open the page, browse the tables and compare models. Use the filters to narrow the view and open the task documentation to see what each task measures. The page takes no commands beyond starting the container.

Example: You are choosing a reranker for a search tool that handles several languages. You open the leaderboard, filter to reranker systems, and compare the listed models, such as bge-reranker-v2-m3, on the tasks that match your use.

For newcomers and experts: If you are new, a retrieval benchmark checks how well a system finds the right documents for a question. Open the page and start with the model comparison view.

If you know the field, BM25 and four neural families share one benchmark here: dense and sparse retrievers, rerankers and late-interaction models. The viewer reads a DuckDB database fetched at startup, so the numbers follow the latest published results.

The container starts
The app fetches the leaderboard database
Tables and comparisons are built from it
You filter and compare in the browser

Good to know:

• License: mit
• Cost to run: none named
• Data it reaches: it downloads the current leaderboard database from Hugging Face when it starts

On Hugging Face, 13 November 2026

• Likes: 27
• Built with: docker
• License: mit

Links

• Space: https://huggingface.co/spaces/hakari-bench/leaderboard
• Files: https://huggingface.co/spaces/hakari-bench/leaderboard/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.