Hugging Face ZeroGPU Free GPU Access on Spaces

Overview

ZeroGPU is Hugging Face's shared GPU infrastructure for Spaces. Instead of holding a dedicated GPU, a Space requests a GPU slice when a decorated function runs and releases it immediately afterwards. All users can run existing ZeroGPU Spaces for free, and free personal accounts can host their own.

Free Tier Quotas & Limits

• Free account: 5 minutes of GPU quota per day, medium queue priority
• Unauthenticated visitors: 2 minutes per day
• PRO account: 40 minutes per day with highest priority; Team organization member: 40 minutes; Enterprise: 60 minutes
• Included daily quota resets 24 hours after your first GPU usage
• Default function runtime is 60 seconds; extend with @spaces.GPU(duration=...)
• Hosting: free personal accounts in good standing (verified email, account older than 30 days) can host up to 2 ZeroGPU Spaces
• PRO accounts can host up to 10; Team and Enterprise organizations up to 50
• PRO, Team, and Enterprise users can extend beyond the daily quota with pre-paid credits at $1 per 10 minutes of GPU time

GPU Hardware

• large (default): half NVIDIA RTX Pro 6000 Blackwell with 48 GB VRAM, 1x quota cost
• xlarge: full NVIDIA RTX Pro 6000 Blackwell with 96 GB VRAM, 2x quota cost

Requirements

• ZeroGPU Spaces are compatible only with the Gradio SDK (version 4 or newer)
• PyTorch 2.8.0 or later; Python 3.10 or 3.12
• Decorate GPU-dependent functions with @spaces.GPU and place models on CUDA at module level
• torch.compile is not supported; use PyTorch ahead-of-time compilation instead

How to Get Started

Create a free Hugging Face account, then either use the curated list of existing ZeroGPU Spaces or create a Gradio Space and select ZeroGPU as its hardware.