CogVideoX 5B
A solid pick for self-hosting and full data control.
Overview
CogVideoX-5B is an open-weight video generation model from Zhipu AI, released under an Apache 2.0 licence that permits commercial use. It produces coherent short clips from detailed text prompts and, in its quantised form, runs on as little as 4.4GB of VRAM, making it practical on a single consumer GPU. The catch: prompts must be written in English and capped at 226 tokens, and generation stays slow, up to 90 seconds per clip on an H100. Because the weights are open, it can be self-hosted entirely within Europe, keeping data on-premise.
Skill profile
Not disclosed
Strengths
- Open-source and self-hostable
Limitations
- API pricing not disclosed
Who is it for
- you want to control cost or self-host
- you have GDPR constraints
Ideal use cases
- Short clip prototyping
- On-prem video generation
- Internal creative demos
- Generative video R&D
- Quick marketing content
Access & availability
Key specifications
Privacy
Run it locally
| Quantization | Disk | RAM / VRAM | Typical hardware |
|---|---|---|---|
| Q4 · recommended | 2.8 GB | 5 GB | Any recent PC/Mac |
| Q8 · balanced | 5.4 GB | 8 GB | 16GB PC / M1+ Mac / 8GB GPU |
| FP16 · max quality | 10 GB | 13 GB | 32GB PC / 24GB Mac / RTX 4070 Ti+ |
Estimates for a moderate context. Long contexts need more RAM (KV cache).
Deploy
Copy-ready commands generated from this card. Adjust context length and GPU count to your hardware.
OpenAI-compatible server for production on NVIDIA GPUs.
pip install vllm
vllm serve THUDM/CogVideoX-5b \
--max-model-len 32768 \
--tensor-parallel-size 1 \
--dtype autoAdvanced data · for expertsArchitecture, modalities, detailed cost, full benchmarks▾
| Arena Elo | — |
| MMLU | — |
| GPQA | — |
| HumanEval | — |
| SWE-Bench | — |
| MATH | — |