fshy

Home / Hugging Face

Hugging Face ZeroGPU Spaces (public Spaces called through the Gradio client)

huggingface.co/docs/hub/en/spaces-zerogpuInference via hosted demos, GPU quotainference/huggingface-zerogpu-spaces-apiJSONcomparereport an errorprovider JSON

Free allowance
ZeroGPU Spaces are available to use for free to all users: 2 minutes of GPU a day unauthenticated and 5 minutes a day on a free account, reset exactly 24 hours after the first GPU usage
Verdict
caveatusable, with a named limit or an unprobed read path
Cost
free tier
Identity
anonymousno card, no phone
Binding limit
volume-quota2 minutes of GPU a day unauthenticated (queue priority Low) and 5 minutes a day on a free account (Medium); PRO is 40 minutes
Evidence
documented6 sources; a first-party page was read and the sentence quoted
Freshness
stale on 2026-11-29checked 2026-10-29 (live); stale on 2026-11-29
Free tierZeroGPU Spaces are available to use for free to all users: 2 minutes of GPU a day unauthenticated and 5 minutes a day on a free account, reset exactly 24 hours after the first GPU usage
unit minutes; metered on wall-clock; the model served is whatever the Space's author runs; the quota counts GPU time of the function call, and the xlarge size costs twice the quota
Pricenone stated
Limits
  • volume-quota (binding) 2 minutes of GPU a day unauthenticated (queue priority Low) and 5 minutes a day on a free account (Medium); PRO is 40 minutes source
  • request-rate a public Space used as an API may be rate limited by Hugging Face if too many requests are made; no number is published source
Bindingvolume-quota minutes of GPU a day bind long before any request count does
Identityanonymous the Unauthenticated tier needs no account and gets 2 minutes a day; a free Hugging Face account is the 5-minute tier. What a free account signup requires is unverified on the pages read
Egressnot published
Interfacerest the Gradio API of a public Space, called with gradio_client.Client(space_id, token=...).predict()
Toolinggradio_client (pip install --upgrade gradio_client); huggingface_hub for the token
Composes withthe hosting side of the same GPU pool is recorded at compute/huggingface-spaces-zerogpu; duplicating a Space to a private one moves the cost to the caller's account
Reliabilityquota left lowers queue priority, and whether a gradio_client call made with a token is counted against that account's quota is not stated on the pages read
NotesWhether the Gradio API path draws on the same daily quota as the web page is unverified.
Evidence notedocumented the daily quota table is quoted from the vendor documentation; no Space was called
Parsed numbers
  • volume-quota 2 minutes
Evidence 6
datesourcewhat it proved
2026-10-28huggingface.co/docs/hub/en/spaces-zerogpuZeroGPU Spaces are free to use for all users
2026-10-28huggingface.co/docs/hub/en/spaces-zerogpuunauthenticated use gets 2 minutes of GPU a day at low queue priority
2026-10-28huggingface.co/docs/hub/en/spaces-zerogpua free account gets 5 minutes of GPU a day at medium queue priority
2026-10-28huggingface.co/docs/hub/en/spaces-zerogputhe daily quota resets 24 hours after the first GPU use
2026-10-28www.gradio.app/guides/getting-started-with-the-python-clienta public Space can be used as an API through the Gradio client and may be rate limited
2026-10-28www.gradio.app/guides/getting-started-with-the-python-clientthe Gradio client takes a Hugging Face token with the token parameter
Unknown values 3
  • $.egress.metered egress metering was not published
  • $.per_object_cap.stated the field does not apply to this kind of offer
  • $.retention.stated the field does not apply to this kind of offer