whatmodel.sh / video / CogVideoX1.5-5B

CogVideoX1.5-5B

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

CogVideoX1.5-5B is an open-weights video model from Zhipu AI (THUDM), released November 2024. Its published size is 5.57B parameters. Licence: cogvideox.

Model information sources

· Weights licence ↗ · Parameter count ↗

Diffusion transformer only: HF Hub safetensors total for the official repo (pinned revision), 5,570,676,352 = the three transformer/ shards. The T5 text encoder and the VAE are not included. Observed 2026-10-07.

36.1±2.4#76 of 93 video modelsfrom 6 benchmark results across 5 sources

$ how it compares

The video models ranked either side of CogVideoX1.5-5B, out of 139 with a score, then more from Zhipu AI (THUDM).

#modelscorewhereRAMreleased
111MAGI-1 4.5B Sand AI37±5comparison only2025-04
112LTX-2.3 Lightricks36±3api $0.040/sec +local~34GB2026-03
113CogVideoX1.5-5B Zhipu AI (THUDM)36±2local~38GB2024-11
114CogVideoX-2B Zhipu AI (THUDM)36±2local~5.3GB2024-08
115Pika 2.2 Pika36±2api $0.040/sec2025-02
more from Zhipu AI (THUDM)
85CogVideoX-5B-I2V Zhipu AI (THUDM)46±4local~13GB2024-09
91CogVideoX-5B Zhipu AI (THUDM)44±2api price unknown +local~13GB2024-08

see all video models · find the ones that fit your machine

$ what it's good at

text to video▆▄▃▄▆▅█▃▂▂▂▄ #53 of 68 · 85% of its score · 5 tests

Trails Pika 2.2, CogVideoX-2B here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
AIGVE-60K human perception MOS (LOVE)50.6% ✎▄▄·█··█▄▄▄▄▄ beats 6/10easy · avg 73%▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%at its level
AIGVE-60K human text-video correspondence MOS (LOVE)49.7% ✎▄▄·█▄█▄··▄▄▄ beats 5/10easy · avg 70%▇▇▇▇▇▇▇▇ 19%at its level
VideoVerse overall (Gemini 2.5 Pro judge)894.05 / 1608 ✎█·▄··▄·····█ beats 0/5easy · avg 1103.17 / 1608▇▇▇▇▇▇ 14%at its level
VBench T2V total (long-prompt version, SkyReels-V2 runs)80.3% ✎█···█···█··█ beats 0/3very easy · avg 88%▇▇▇▇▇ 12%typical
Sci-VBench testmini (expert human rating)38.6% ✎▅·▃█▃·▅▃▅▃▃▃ beats 1/14hard · avg 44%▇▇▇ 6.4%blunt
other tests1 not in a task · 13% of its score
testresultvs the others in its rundifficultycountswhy
VideoArgus-Bench T2V (rubric-grounded score)75.2% ✎█▄·▄█▄▄····▄ beats 1/8easy · avg 82%▇▇▇▇▇ 13%blunt · at its level

$ where to run it

Other ways to run it

  1. Run it as published · needs ~37.8 GB, NVIDIA only · from huggingface.co/zai-org
    uv run --with torch --with git+https://github.com/huggingface/diffusers --with transformers --with accelerate --with imageio-ffmpeg --with sentencepiece python -c "import torch; from diffusers import CogVideoXPipeline; from diffusers.utils import export_to_video; p=CogVideoXPipeline.from_pretrained('zai-org/CogVideoX1.5-5B', torch_dtype=torch.bfloat16); p.enable_sequential_cpu_offload(); p.vae.enable_tiling(); p.vae.enable_slicing(); export_to_video(p(prompt='A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest.', num_videos_per_prompt=1, num_inference_steps=50, num_frames=81, guidance_scale=6, generator=torch.Generator(device='cuda').manual_seed(42)).frames[0], 'output.mp4', fps=8)"

Weights and documentation: huggingface.co/zai-org ↗

$ how the score was worked out · LOVE: Benchmarking and Evalua… 53%, VideoVerse: Does Your T2V Gen… 14%, VideoArgus: Agentic Rubric-Gr… 13% and 2 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

sourceresultsshare of the scorelicence
LOVE: Benchmarking and Evaluating Text-to-Video Generation and Video-to-Text Interpretation (AIGVE-60K)2▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 53%CC-BY-4.0
VideoVerse: Does Your T2V Generator Have World Model Capability to Synthesize Videos?1▇▇▇▇ 14%CC-BY-4.0
VideoArgus: Agentic Rubric-Grounded Unified Evaluation for Video Generation and Editing (arXiv:2608.05485v1)1▇▇▇▇ 13%CC-BY-4.0
SkyReels-V2: Infinite-length Film Generative Model1▇▇▇ 12%CC-BY-4.0
Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains1▇▇ 6%CC-BY-4.0
prior · pull toward the middle▇ 3%
weighted average= 36.1

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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