Spaces (Hugging Face)

Spaces is Hugging Face's hosted demo platform — deploy Gradio / Streamlit / static / Docker apps as live shareable demos. The de-facto standard for ML demo + research artifact distribution in 2026.

Pre-Spaces, researchers + indie devs had to deploy demos themselves — high friction. Spaces lowered the bar: write `app.py` with Gradio or Streamlit, push to a Spaces repo, the demo lives at `huggingface.co/spaces/<user>/<demo>` with autoscaling + free CPU tier (paid GPU). Use cases: paper demos (let reviewers try the model), prototype ML apps without infra, A/B test models with non-technical stakeholders, share work with the ML community. Variants: ZeroGPU (free GPU bursts), Dedicated GPU (always-on), Docker Spaces (custom Dockerfile). By 2026 Spaces hosts hundreds of thousands of live ML demos — research community + indie devs default to it.

When to use spaces (hugging face)

• Paper / model demos.
• Prototype ML apps without infra.

Common mistakes

• Production traffic on free CPU tier — gets queued / killed on load.

FAQ

Related terms

• Model zoo — A model zoo is the platform-hosted collection of pre-trained models you can browse, download, fine-tune, or serve — Hugging Face Hub (the dominant 2026 model zoo), Replicate, Civitai, ModelScope are leaders.
• Live preview — Live preview is the IDE / chat feature where rendered output (web page, React component, Mermaid diagram, SVG, plot) updates in real time as the user or LLM edits the source — Claude Artifacts, v0, Bolt, Lovable, Storybook are 2026 examples.

Sources

• HF Spaces docs

Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/spaces.md.