Hugging Face Spaces

Hugging Face Spaces is a repository-based service for creating, deploying, and sharing machine-learning demos and applications on the Hugging Face Hub. A Space stores its source in a Git repository and can run a Gradio app, Docker container, or static HTML/JavaScript app. It is useful for publishing interactive model experiences, prototyping, teaching, and collaborating with the ML community.

✨ Key Features

• Create ML demos — Start a Space from the Hub and choose Gradio, Docker, or static HTML as the application approach.
• Git-based development — Store code and configuration in a repository; each pushed commit can rebuild and restart the Space.
• Public, protected, or private visibility — Share a running app publicly, keep source code protected while exposing an embed URL, or restrict both source and app to collaborators.
• Model and dataset integration — Use Hugging Face Hub models, datasets, and files from an application and preload selected Hub assets during build.
• CPU and GPU hardware — CPU Basic has no hourly charge, but creating compute-based Gradio/Docker Spaces requires personal PRO or organization Team/Enterprise. Eligible free personal accounts can host up to two ZeroGPU Gradio Spaces; Static Spaces remain free.
• Embeds and custom domains — Expose public or protected Spaces through an embed URL and configure a custom domain where supported.
• Secrets and variables — Configure runtime secrets and environment variables for API keys and application settings.
• Persistent data through Buckets — Attach Storage Buckets as volumes when an application needs data to survive restarts; the default Space disk is ephemeral.
• Collaboration and portfolio sharing — Work with an organization or collaborators, showcase projects, and share demos with stakeholders.
• Runtime lifecycle controls — Pause Spaces, use sleep behavior on free hardware, and control cost by changing hardware or stopping execution.

💳 Pricing

• Static Spaces — Free for everyone; static HTML/JavaScript apps do not require compute.
• Free personal ZeroGPU — Eligible personal accounts in good standing can host up to 2 Gradio Spaces on ZeroGPU, subject to quota and queue rules.
• PRO — $9/month; includes the ability to host ZeroGPU, Gradio, and Docker Spaces, 8× ZeroGPU quota, higher queue priority, Spaces Dev Mode, and personal Hub benefits.
• Team — $20/user/month; adds organization collaboration, SSO, audit logs, resource groups, access controls, and Gradio/Docker Spaces with advanced compute options.
• Enterprise — Custom pricing on the dedicated Team/Enterprise page; the general pricing page separately displays $50/month without establishing the applicable contract. Confirm a current sales quote rather than treating $50 as a universal fee; adds custom onboarding, higher storage/bandwidth/API limits, SCIM, advanced security, managed billing, compliance processes, and dedicated support.

Compute beyond free allowances is billed by the minute while the Space is Starting or Running. Published examples include CPU Upgrade at $0.03/hour, T4 small at $0.40/hour, L4 at $0.80/hour, and larger GPU options at higher rates. Hub storage, Buckets, taxes, and usage can add cost; verify the live pricing page and account eligibility before purchase.

⚖️ Alternatives Compared

🎯 Use Cases

• Publishing a model demo — Wrap a model in Gradio and share an interactive public app with a research group or customer.
• Prototyping an AI product — Test a model-powered interface before investing in a separate application deployment.
• Teaching and workshops — Give learners a runnable ML example that can be inspected and forked from a repository.
• Showcasing research — Demonstrate a paper, model, dataset, or benchmark through a shareable web application.
• Running GPU experiments — Select suitable hardware for inference or demos and account for minute-based compute billing.
• Building an internal tool — Use protected or private visibility, secrets, and organization collaboration for a team application.
• Connecting Hub assets — Load models or datasets from the Hub and keep application code, configuration, and revisions together.
• Persisting application data — Attach a Storage Bucket when ephemeral Space disk is not sufficient for runtime data.