DeepSeek-V3.2
DeepSeek-V3.2 is an open-weight language model from DeepSeek, released on December 1, 2025. It is a mixture-of-experts model with 671B total parameters, of which 37B are active for each token. The model takes text input.
- Developer
- DeepSeek
- Released
- December 1, 2025
- Total parameters
- 671B
- Active parameters per token
- 37B
- Architecture
- MoE
- Input
- Text
- License
- MIT
- Commercial use
- Yes
- Official weights, GB
- 1,370.8 GB
- Hugging Face repository
- huggingface.co
- Checked on
The weights are released under the MIT license, which allows commercial use.
The official weights on Hugging Face take about 1,371 GB. Running the model needs at least that much memory across GPUs and system RAM, plus room for the context cache; quantized versions need less.
- Total parameters, billions
- 671B
- Active parameters, billions
- 37B
- License type
- MIT
- License conditions
- MIT: keep the copyright and permission notice.
- Weights precision
- FP8 + BF16
- License text
- huggingface.co
- Sources
Page URL Hugging Face model card https://huggingface.co/deepseek-ai/DeepSeek-V3.2/raw/main/README.md Hugging Face weights index https://huggingface.co/deepseek-ai/DeepSeek-V3.2/raw/main/model.safetensors.index.json DeepSeek API docs: release notes https://api-docs.deepseek.com/news/news251201 Hugging Face model card https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro/raw/main/README.md