Open VLM Video Leaderboard

Open VLM Video Leaderboard

A leaderboard of how vision-language models score on video understanding benchmarks, with an overall ranking and a detailed board for each benchmark. For researchers and engineers choosing a model.

Made by opencompass 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: Open VLM Video Leaderboard is a results page for video understanding. It collects scores that the VLMEvalKit toolkit produced for vision-language models. One board ranks models overall across a chosen set of benchmarks, and further boards list overall and fine-grained scores for each single benchmark.

Who it suits: It suits people who need to compare models on video tasks and would rather read measured scores than vendor claims.

Title: Who might use this leaderboard
Industry | Who uses it | How they use it | Example
Model research | Researchers | Compare video-capable models on shared benchmarks | Check where a new model lands in the overall ranking
AI engineering | Machine learning engineers | Shortlist a model for a video feature | Open the benchmark closest to their task
Academic writing | Paper authors | Look up scores from one evaluation source | Read a benchmark's fine-grained scores

If you want to evaluate your own model, you can skip it, because the Space shows results and does not run evaluations.

Where to use it: You can read it online or run a copy yourself.

• Web browser: the Space's page on Hugging Face, with nothing to install
• Windows, macOS or Linux: a local copy that runs with Python
• Hugging Face Spaces: where the hosted board runs, on the cpu-basic hardware tier

Title: What the leaderboard is good for
Use case | Who | What they get back
Rank models overall | Researchers | Total scores across a selected group of video benchmarks
Study one benchmark | Engineers | Overall and fine-grained scores for that benchmark alone
Build a reading list | Students | A list of models worth reading about, ordered by score

How to set it up: The hosted board needs nothing. A local copy needs Python and pip.

• Python, with pip
• The Python packages that requirements.txt lists

1. Install the huggingface_hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio 4.44.0.
5. Install the packages from requirements.txt.
6. Start the app with Python.

pip install -U huggingface_hub
hf download opencompass/openvlm_video_leaderboard --repo-type space --local-dir openvlm_video_leaderboard
cd openvlm_video_leaderboard
pip install gradio==4.44.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space online, or start your copy with the command below. Then read the overall board first and move to a single benchmark's board when you need more detail.

python app.py

Title: The boards you will meet
Tool | What it does
Overall leaderboard | Ranks models by total score across carefully selected video benchmarks
Benchmark leaderboards | Show overall and fine-grained scores for one video benchmark each
app.py | Starts the Gradio page when you run it with Python

Example: You wonder how a certain model does at video understanding. You open the overall board to see its total score against the others, then open one benchmark's board to see the fine-grained scores behind that total.

For newcomers and experts: If you are new to this, open the overall board and scan the ranking. Then pick one benchmark and see whether the order changes.

If you know the field, the scores come from VLMEvalKit runs, so they share one evaluation setup. The overall board uses a curated benchmark set, and the per-benchmark boards expose the finer breakdown. The page asks you to cite the VLMEvalKit report when the results help your research.

VLMEvalKit evaluates the models
Scores are collected into the Space
The overall board ranks them
Benchmark boards show the detail

Good to know: This Space reports results and does not evaluate models itself.

• Scope: it covers video understanding benchmarks only
• Citation: the page asks you to cite the VLMEvalKit report if the resource helps your research
• License: apache-2
• Cost to run: none named
• Data it reaches: the leaderboard data the app shows, plus the Python packages pip installs from requirements.txt

On Hugging Face, 13 November 2026

• Likes: 136
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard
• Files: https://huggingface.co/spaces/opencompass/openvlm_video_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.