Communauté IA open source hébergeant modèles, jeux de données et applications de démo Spaces
Modèles IA Modèles et poids IA huggingface.co
What is Hugging Face
Hugging Face is the world's largest community for open-source AI models and datasets. Based in New York, it was founded in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond and Thomas Wolf and named after the 🤗 emoji. It began as a chatbot app for teenagers, then pivoted to a machine learning platform after open-sourcing the model behind it. It also develops the widely used open-source Transformers library. In August 2026 Hugging Face agreed to be acquired by Nvidia.
What it can do
- Model and dataset Hub: almost every major open model is published here and can be downloaded for free
- Spaces for deploying AI demos with Gradio, Streamlit, static pages or Docker
- Open-source libraries such as Transformers, Diffusers, Datasets and PEFT, standard tools for training and running models
- Inference: Inference Providers gives one interface to many inference providers, and Inference Endpoints deploys dedicated inference
- Community: model discussions, leaderboards, papers and blogs
Pricing
Browsing and downloading open models is free. Official paid plans:
- PRO (personal): US$9 a month, 10x private storage, US$2 of inference credits a month and a higher ZeroGPU quota
- Team: US$20 per user a month, with SSO, audit logs and storage region control
- Enterprise: from US$50 per user a month, with SCIM user management and the highest storage and rate limits
Spaces hardware and dedicated inference are billed by time, with dedicated inference from US$0.033 an hour.
Who it's for
- Developers and researchers who want to download open models to deploy or fine-tune
- Anyone who wants to put up an AI demo for others to try quickly
- Companies that need private model repositories and team collaboration
Good to know
- Licences differ by model; some don't allow commercial use or have revenue thresholds, so read the licence on the model page before downloading.
- Access from mainland China can be unreliable; Alibaba's ModelScope is an alternative.
Similar tools
Compare ModelScope and Replicate; for running models locally, see Ollama.