whatmodel.sh / video / LongCat-Video
LongCat-Video
↗ share
+ contribute
LongCat-Video is an open-weights video model from Meituan, released October 2025. Its published size is 13.6B parameters. Licence: mit.
Model information sources
· Weights licence ↗ · Parameter count ↗
Card table also gives '# Total Params 13.6B / # Activated Params 13.6B' (dense). Covers the video generation model.
$ how it compares
The video models ranked either side of LongCat-Video, out of 139 with a score.
| # | model | score | where |
|---|---|---|---|
| 61 |  PixVerse 5.5 | 52±10 | api $0.060/sec |
| 62 |  Kling 1.6 (Pro mode) | 52±4 | comparison only |
| 63 | LongCat-Video | 51±2 | api $0.040/sec +local |
| 64 |  Kandinsky 6.0 Video Lite | 51±7 | local |
| 65 |  Seaweed-7B | 50±6 | comparison only |
$ what it's good at
text to video▆▄▃▄▆▅█▃▂▂▂▄ #33 of 68 * · 7.5% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Sci-VBench testmini (expert human rating) | 45.4% ✎ | ▅·▃█▃·▅▃▅▃▃▃ beats 5/14 | 7.5% |
image to video▁▂·▂▅█▅▄▅▆▃▄ #40 of 50 * · 67% of its score · 1 test
Beats HunyuanVideo 1.5 here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| WorldReasonBench human Elo (image+text-to-video world-state prediction) | 904.0 ✎ | ▃▃··█··▃▃▃·▃ beats 2/8 | 67% |
other tests2 not in a task · 23% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MindWorldBench intention accuracy (I2V mental-state reasoning) | 22.2% ✎ | ▅·▃▅·█·▃···▃ beats 3/9 | 14% |
| VideoArgus-Bench T2V (rubric-grounded score) | 81.1% ✎ | █▄·▄█▄▄····▄ beats 3/8 | 9.1% |
$ where to run it
Through an API
api · fal.ai · $0.040/sec · fal.ai ↗ · price sourceUSD per generated second of 720p video on fal.ai: fal states 'Your request will cost $0.04 per generated second of video. Generated seconds are calculated at 30 frames per second.' Product: fal LongCat Video text-to-video (720p/30fps) endpoint. API model: fal-ai/longcat-video/text-to-video/720p (the base model; the distilled endpoints are separate). Settings: resolution=720p, 30 fps. Price basis: per-second rate printed by fal. 480p endpoint $0.025 per generated second (counted at 15 fps). Currency: USD. Not a Meituan direct tariff. fal.ai model page, observed 2026-09-26.
Other ways to run it
- Run it as published · needs ~100.4 GB, NVIDIA only · from github.com/meituan-longcat
git clone https://github.com/meituan-longcat/LongCat-Video && cd LongCat-Video && pip install torch==2.6.0+cu124 torchvision==0.21.0+cu124 torchaudio==2.6.0 --index-url https://download.pytorch.org/whl/cu124 && pip install ninja psutil packaging && pip install flash_attn==2.7.4.post1 && pip install -r requirements.txt && pip install "huggingface_hub[cli]" && huggingface-cli download meituan-longcat/LongCat-Video --local-dir ./weights/LongCat-Video && torchrun run_demo_text_to_video.py --checkpoint_dir=./weights/LongCat-Video --enable_compile
Weights and documentation: huggingface.co/meituan-longcat ↗
$ how the score was worked out · WorldReasonBench: Human-Align… 67%, MindWorldBench: Evaluating Me… 14%, VideoArgus: Agentic Rubric-Gr… 9% and 1 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json