CogVideo
Z.aiCogVideo is Z.ai's line of open text-to-video and image-to-video diffusion models. Its current generation, CogVideoX1.5 (November 2024), generates 5- or 10-second clips at up to 1360x768 from a prompt or a starting image; the earlier CogVideoX-2B and 5B models remain available.
No CogVideo model has followed CogVideoX1.5, and the repository was last updated in November 2025. CogVideoX-2B is Apache 2.0; every 5B checkpoint, including both CogVideoX1.5 models, ships under the CogVideoX License, which allows free commercial use after registering with Z.ai, up to one million visits a month.
Openness
3 medium confidence- weights
- open (CogVideoX1.5-5B and the earlier CogVideoX checkpoints download from the Hub without a gate)
- data
- described (the CogVideoX paper describes its text-video data processing and captioning pipeline; no training set is released)
- code
- partial (inference code plus LoRA and full fine-tuning scripts for users; the pretraining code is not published)
- license
- CogVideoX-License (the CogVideoX1.5 and 5B weights: free for academic research; commercial use is free after registration, up to one million visits a month, with a separate license above that)+Apache-2.0 (the repository code and the CogVideoX-2B weights)
CogVideoX1.5's weights are free for research, and free for commercial use once a company registers with Z.ai, up to a million visits a month. The code is Apache 2.0. Fine-tuning scripts are published but the pretraining code is not, and the training videos are described but not released.
- https://arxiv.org/abs/2408.06072 recorded 2026-11-12
Paper abstract: "we develop an effective text-video data processing pipeline that includes various data preprocessing strategies and a video captioning method"; the weights of the VAE, caption model and CogVideoX are public, the training data is not.
- https://huggingface.co/api/models?author=zai-org&search=CogVideo&sort=downloads&direction=-1&limit=100 recorded 2026-11-12
Listing of the zai-org CogVideo repositories: CogVideoX-2b, CogVideoX-5b, CogVideoX-5b-I2V, CogVideoX1.5-5B and CogVideoX1.5-5B-I2V are public; none is newer than November 2024.
- https://huggingface.co/api/models/zai-org/CogVideoX1.5-5B?expand[]=downloads&expand[]=likes&expand[]=cardData&expand[]=gated&expand[]=lastModified&expand[]=createdAt&expand[]=siblings recorded 2026-11-12
Hugging Face API record for zai-org/CogVideoX1.5-5B: license other, gated false, transformer, VAE and text-encoder safetensors.
- https://huggingface.co/zai-org/CogVideoX-2b/raw/main/LICENSE recorded 2026-11-12
CogVideoX-2B LICENSE body: Apache License, Version 2.0.
- https://huggingface.co/zai-org/CogVideoX1.5-5B/raw/main/LICENSE recorded 2026-11-12
License body: "The CogVideoX License"; "Users who wish to use the models for commercial purposes must register and obtain a basic commercial license"; "the number of service users (visits) for your commercial activities must not exceed 1 million visits per month."
- https://ungh.cc/repos/zai-org/CogVideo/files/main recorded 2026-11-12
Repository file list: inference scripts and a finetune/ directory with LoRA and SFT trainers for CogVideoX and CogVideoX1.5; no pretraining pipeline.
- https://ungh.cc/repos/zai-org/CogVideo/files/main/README.md recorded 2026-11-12
README Model-License: "The code in this repository is released under the Apache 2.0 License"; the CogVideoX-2B model is Apache 2.0 and the 5B models use the CogVideoX LICENSE; news: "2024/11/08: We have released the CogVideoX1.5 model."
Adoption
2 high confidenceAdoption is measured as Hugging Face downloads summed across the five zai-org CogVideoX checkpoints; the older CogVideoX-2B and 5B account for almost all of them.
- https://huggingface.co/api/models?author=zai-org&search=CogVideo&sort=downloads&direction=-1&limit=100 recorded 2026-11-12
Listing sorted by downloads: CogVideoX-2b 14559, CogVideoX-5b 8554, CogVideoX-5b-I2V 6699, CogVideoX1.5-5B-I2V 620, CogVideoX1.5-5B 599; 31,031 in total.
Capability
1 medium confidenceCogVideoX does not appear on the public video arenas, which now list only current models, and sits level with Open-Sora, another open video line with no placement there.
- https://artificialanalysis.ai/video/leaderboard/image-to-video recorded 2026-11-12
The image-to-video page serves a with-audio list of 35 models and a without-audio list of 40; neither has a CogVideoX entry.
- https://artificialanalysis.ai/video/leaderboard/text-to-video recorded 2026-11-12
The text-to-video page serves a with-audio list of 34 models and a without-audio list of 39; neither has a CogVideoX entry.
- https://huggingface.co/zai-org/CogVideoX1.5-5B/raw/main/README.md recorded 2026-11-12
CogVideoX1.5-5B card: video resolution 1360 * 768, 5 or 10 seconds at 16 frames per second.
Verified 2026-11-12