Skip to main content
Hugging Face Inference API logo

Is Hugging Face Inference API Down?

No — Hugging Face Inference API is up
Hugging Face Inference API's status page
Not available · we could not read it
FlareWarden checks
Website reachable · 8 of 8 regions · 164 ms

Right now

We could not read Hugging Face Inference API's status page, so the verdict rests on our checks. Check it yourself How we decide

FlareWarden checks

huggingface.co from 8 regions · when you open this page (at most every 30 minutes), otherwise every 6 hours

Reachable
Regions up
8 of 8
Response
164 ms
Checked

By region

  • Ashburn reachable, 43 ms
  • Chicago reachable, 93 ms
  • Johannesburg reachable, 346 ms
  • London reachable, 114 ms
  • Los Angeles reachable, 113 ms
  • San Jose reachable, 114 ms
  • Singapore reachable, 280 ms
  • Tokyo reachable, 214 ms
Timing by region
Connection timing for each region that loaded the page
RegionDNSTCPTLSTTFB
Ashburn22 ms1 ms5 ms43 ms
Chicago51 ms1 ms4 ms92 ms
Johannesburg74 ms0 ms3 ms346 ms
London20 ms2 ms3 ms114 ms
Los Angeles30 ms1 ms4 ms113 ms
San Jose31 ms1 ms4 ms114 ms
Singapore45 ms1 ms3 ms280 ms
Tokyo38 ms1 ms4 ms214 ms
We check that huggingface.co loads, not every feature behind it. Test it yourself

History

Hugging Face Inference API website availability

Measured by FlareWarden · huggingface.co

Last 7 days
100%
Last 30 days
100%
Last 90 days
100%
Avg response, 30 days
181ms

Last 30 days

70-day clean streak
Sep 25: 100.00% uptime, 32 checks
Sep 26: 100.00% uptime, 32 checks
Sep 27: 100.00% uptime, 31 checks
Sep 28: 100.00% uptime, 32 checks
Sep 29: 100.00% uptime, 32 checks
Sep 30: 100.00% uptime, 32 checks
Oct 1: 100.00% uptime, 31 checks
Oct 2: 100.00% uptime, 32 checks
Oct 3: 100.00% uptime, 32 checks
Oct 4: 100.00% uptime, 32 checks
Oct 5: 100.00% uptime, 32 checks
Oct 6: 100.00% uptime, 32 checks
Oct 7: 100.00% uptime, 32 checks
Oct 8: 100.00% uptime, 32 checks
Oct 9: 100.00% uptime, 32 checks
Oct 10: 100.00% uptime, 32 checks
Oct 11: 100.00% uptime, 32 checks
Oct 12: 100.00% uptime, 32 checks
Oct 13: 100.00% uptime, 32 checks
Oct 14: 100.00% uptime, 32 checks
Oct 15: 100.00% uptime, 32 checks
Oct 16: 100.00% uptime, 31 checks
Oct 17: 100.00% uptime, 32 checks
Oct 18: 100.00% uptime, 32 checks
Oct 19: 100.00% uptime, 32 checks
Oct 20: 100.00% uptime, 32 checks
Oct 21: 100.00% uptime, 32 checks
Oct 22: 100.00% uptime, 32 checks
Oct 23: 100.00% uptime, 32 checks
Oct 24: 100.00% uptime, 24 checks
Sep 25 Today
No downtime Partial Downtime Not measurable No data

Measured by FlareWarden's own checks of huggingface.co from multiple regions every 6 hours. Hugging Face Inference API's status page sets the verdict above when it reports trouble, but never these figures. Percentages count only checks that returned an availability answer — 70 days measured so far. A dash means that window does not yet hold enough measured days to publish a figure.

What Hugging Face Inference API does

Hugging Face hosts the model and dataset repository most machine learning work pulls from, along with hosted inference endpoints and Spaces demos. Training scripts, build pipelines, and production applications download weights from it at runtime, which makes it a build-time and deploy-time dependency for teams who never open the website themselves.

What an outage looks like

Model downloads fail partway with connection or 5xx errors, and code loading a pretrained model hangs or raises a repository-not-found error for models that plainly exist. Git pushes and pulls of large model files stall separately from the website. Spaces demos show a proxy error, and hosted Inference Endpoints time out.

What to do about it

Check status.huggingface.co, which separates the Hub, git hosting, Spaces, the CDN, and Inference Endpoints, because these fail independently and the component you depend on may be fine. Locally cached models keep working, so setting HF_HUB_OFFLINE stops libraries revalidating against the network mid-incident and turns long timeouts into immediate cache hits.

Is it down for everyone, or just you?

If this page says Hugging Face Inference API is up but it is not loading for you, the problem is between you and them. Run a check against any URL from up to 17 cities across three regions to find out where it breaks.

Test it yourself

Related services

AWS
AWS
Cloudflare
Cloudflare
GitHub
GitHub
Vercel
Vercel

Hugging Face Inference API outage FAQ

Will my running application break if Hugging Face goes down?
Only if it fetches from the Hub at runtime. Applications that downloaded weights at build time and hold them in a local cache or image layer keep serving normally. Those loading a pretrained model on every cold start will fail once the cache is empty, which is why baking models into the image is common practice.
Is the Hub down or just Inference Endpoints?
They are tracked as separate components and often fail apart. The status page lists the Hub, git hosting, Spaces, the CDN, and Inference Endpoints on their own lines. A CDN fault can break model downloads while the website loads fine, so match the component to the call your code is actually making.
Can I keep working offline during an outage?
Yes, if the models are already cached. Setting HF_HUB_OFFLINE to 1 makes the libraries use the local cache without contacting the network, which avoids a long timeout on every call. Anything not yet downloaded stays unavailable, so the technique rescues work in progress rather than letting you start something new.
Does a Hugging Face outage affect other model APIs?
No. Commercial model providers run separate infrastructure with their own status pages. A Hugging Face incident affects downloading open models and calling endpoints hosted there. Applications routing to a commercial API for generation but to Hugging Face for tokenisers or embeddings may find one path works and the other does not.

How we measure this

  • We request Hugging Face Inference API's public endpoint every 6 hours from Fly.io regions across six continents — the most recent check ran from 8 of them. When you open this page and our last check is more than 30 minutes old, we check again on the spot; those checks update the verdict, never the availability figures.
  • A region counts as down only when it gets no usable HTTP response. A 403 or 429 means the origin answered and refused us, which we report as blocked, never as an outage.
  • A single failing region is treated as probe noise. Our checks only change their verdict when two consecutive cycles agree.
  • Response times average only the regions that actually served the page, so a timeout never inflates the number.
  • Where Hugging Face Inference API publishes an official status feed we read it too, and the verdict is the worse of what it reports and what our checks see: a major outage of Hugging Face Inference API itself is Down, trouble with one part of it is Degraded, problems at a few of a provider's many locations change nothing, and a security notice that marks nothing down is shown but changes nothing either. A reading counts for 45 minutes; we read the feed again within minutes of any change to its incidents. The availability figures above are our measurement alone.

Get alerted when Hugging Face Inference API goes down

We check Hugging Face Inference API every 6 hours, and again when you open this page if that was more than 30 minutes ago. Your own monitors run as often as every 30 seconds, from up to 17 cities across three regions, and tell you the moment something breaks.

Start Free Monitoring
Free plan available No credit card required