# What is the free ZeroGPU quota for 1 space? **URL:** \ **Category:** Spaces\ **Created:** [August 12, 2026, 11:42am UTC](https://discuss.huggingface.co/t/what-is-the-free-zerogpu-quota-for-1-space/178610 "2026-08-12T11:42:35Z")\ **Posts on this page:** 1\ **Showing post:** 2
**Author:** ![John6666](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/john6666/32/27664_2.png) [@John6666](https://discuss.huggingface.co/u/John6666)\ **Post date:** [August 12, 2026, 9:13pm UTC](https://discuss.huggingface.co/t/what-is-the-free-zerogpu-quota-for-1-space/178610/2 "2026-08-12T21:13:45Z")
Um, I’ve never personally hit a per-Space limit, but **Free accounts do have another limit** , so that might be what you’re seeing: * * * The currently documented ZeroGPU quota for a **Free account is 5 GPU-minutes per day**. The [current ZeroGPU documentation](https://huggingface.co/docs/hub/spaces-zerogpu) describes that as an **account-tier daily quota** , not as “5 minutes for each Space.” It also says the included quota resets exactly 24 hours after your first GPU usage. However, there is another limit which is easy to miss because it is not currently listed in that quota table. In [this April 2026 Forum thread](https://discuss.huggingface.co/t/free-account-zerogpu-quota-issue/175180), `hysts` explained that Free users had a separate limit on **how many times they could run ZeroGPU Spaces per day** , in addition to the time-based quota. When someone asked whether both limits existed, the answer was yes; and [the follow-up answer](https://discuss.huggingface.co/t/free-account-zerogpu-quota-issue/175180/9) said that the request-based limit **at that time was 3 runs per day**. So the model was roughly: ```plaintext Free account ├─ time-based quota: 5 GPU-min/day └─ request/run-based limit: 3/day (according to the April 2026 HF Forum answer) ``` That second limit could explain a situation where your quota meter still seems to have plenty of GPU time left, but you can no longer run a Space. One caveat: I would not treat **3/day** as a permanently documented current value. The official ZeroGPU docs currently show the 5-minute time quota but do not show the request-count limit, and I could not find a newer public HF statement saying whether the April value has since changed or been removed. So the safest wording is: **HF staff said it was 3/day in April; it may still be the limit, but the current docs do not expose that number.** > **Why one particular Space can still appear to have a lower quota** > > There are also a couple of Space-level effects that can make the behavior look like “this Space has a smaller quota,” even if there is no separate daily quota specifically assigned to that Space. > > ### 1. Different GPU duration per request > > ZeroGPU functions use `@spaces.GPU`. According to the [current docs](https://huggingface.co/docs/hub/spaces-zerogpu#duration-management), the default GPU duration is 60 seconds, but a Space author can request a different duration, for example: > > ```python > @spaces.GPU(duration=120) > def generate(...): > ... > > ``` > > So two Spaces can place very different demands on the same remaining account quota. > > Conceptually: > > ```plaintext > Account quota remaining: 70 seconds > > Space A wants about 60 seconds > -> request can fit > > Space B wants about 120 seconds > -> request cannot fit > > ``` > > That can look like Space B has its own smaller allowance, although it is really the same account quota interacting with a larger per-call request. > > ### 2. `xlarge` has a 2× quota cost > > Current ZeroGPU has two GPU sizes: > > | Size | Quota cost | > | --- | --- | > | `large` | 1× | > | `xlarge` | 2× | > > The [official docs](https://huggingface.co/docs/hub/spaces-zerogpu#gpu-size-selection) give the example that a 45-second effective task on `xlarge` consumes 90 seconds of quota. > > There was also a [May 2026 Forum discussion](https://discuss.huggingface.co/t/getting-quota-exceeded-even-though-requested-seconds-is-less-than-whats-left/176021) where `hysts` explained that, after the ZeroGPU hardware migration, some workloads were automatically falling back to `xlarge`. Because `xlarge` costs 2× quota, this made some Spaces appear to consume much more quota than expected. > > So if the symptom is **not** “it stops after about three runs,” these are the next things I would suspect. A simple way to distinguish the cases is: ```plaintext Still have GPU minutes left | +-- already made ~3 ZeroGPU runs today | -> the separate Free request/run limit is a strong candidate | +-- only 1–2 runs, but this particular Space refuses -> per-request duration / xlarge quota cost becomes more interesting ``` Also, in the April thread, `hysts` noted that when the new run limit was introduced, reaching it could temporarily produce the **same kind of quota error** as exhausting GPU time. So the wording of a “quota exceeded” error by itself does not necessarily tell you which limit was reached. So, if by “one Space seems to be lower” you mean **“I still have some of my 5 minutes left, but after roughly three uses it stops working,”** I would first suspect the separate Free-account run limit rather than a special quota assigned to that individual Space. If instead **that one Space fails after only one or two attempts while other ZeroGPU Spaces still work** , then it is probably worth looking at that Space’s requested duration / GPU size or treating it as a Space-specific issue. --- _[View the full topic](https://discuss.huggingface.co/t/what-is-the-free-zerogpu-quota-for-1-space/178610)._