Is Hugging Face Down?
Live Hugging Face status. Auto-refreshes every 2 minutes.
Hugging Face is Operational
Hugging Face is up and running normally. All systems are operational.
Official status page: status.huggingface.co
Component Status
Recent Hugging Face incidents
What to do when Hugging Face is down
Route inference through a sibling gateway
If the Hugging Face Inference API or your dedicated Endpoint is failing but you just need to run an open-weights model, the fastest fix is to send the same model to another inference host. Together, Fireworks, Groq, and OpenRouter all serve popular open models like Llama, Mistral, and Qwen with OpenAI-compatible APIs, so swapping the base URL and key is usually a small change. Check that the target gateway is itself healthy before you cut over.
Fall back to local cache or another model source
If Hub downloads are failing, models and datasets you already pulled stay available in your local cache, so set HF_HUB_OFFLINE=1 to keep training and inference running without network calls. For new weights, Replicate hosts many of the same open models behind a simple API, and the open-weights tracker shows which models are mirrored where. Confirm the alternative is operational before you depend on it.
Get notified when status changes
Instead of refreshing this page during an incident, subscribe to TensorFeed alerts and get a ping the moment Hugging Face flips between operational, degraded, and down. You can also watch the full multi-provider dashboard to see whether an issue is isolated to Hugging Face or part of a wider outage hitting several AI services at once.
Frequently Asked Questions
Is Hugging Face down right now?
The live indicator at the top of this page shows the current state. Green means the Hub, Inference, and Spaces are operational, amber means degraded performance (slow model downloads, queued Spaces, or intermittent Inference errors), and red means a confirmed outage. If you see green here but a specific feature is failing for you, it is more likely a model-specific, repo-specific, or local network issue than a platform-wide outage.
How do I check if Hugging Face is down?
Read the live status panel on this page first, which tracks the Hub, Inference, and Spaces together. For the authoritative per-component breakdown, open the official status page at status.huggingface.co, which lists incidents for the website, Hub APIs, Inference, and Spaces separately. You can also confirm scope quickly by trying a `git clone` or `huggingface-cli download` against a public repo; if that fails too, the problem is on Hugging Face's side rather than yours.
What should I do when Hugging Face is down?
First identify which surface is failing, since the Hub, Inference API or Endpoints, and Spaces can fail independently. If only Inference is down, route the same open model through a sibling inference gateway like Together, Fireworks, Groq, or OpenRouter. If model downloads from the Hub are failing, models you already cached locally (in your `~/.cache/huggingface` directory) keep working offline, so set `HF_HUB_OFFLINE=1` to force the cache. For Spaces outages, there is no quick substitute beyond waiting or self-hosting the underlying code.
How often does Hugging Face go down, and how reliable is it?
Hugging Face is generally reliable for a platform of its scale, and full platform-wide outages are uncommon. The more frequent pattern is partial degradation: slow or rate-limited model downloads during heavy traffic, Inference API timeouts or cold-start delays on less-used models, and Spaces that get stuck building or sleeping. Because the free Inference API is a shared, best-effort tier, intermittent slowness there is normal and not the same as the Hub being down.
Which services and models does Hugging Face actually run?
Hugging Face hosts hundreds of thousands of community and vendor model repos rather than running a single flagship model of its own, so a model failing is often the individual repo or a third-party Inference Provider, not Hugging Face itself. The platform has three distinct surfaces people confuse: the Hub (git-based hosting for models and datasets), Inference (the shared serverless API plus paid dedicated Inference Endpoints), and Spaces (hosted Gradio and Streamlit apps). A common point of confusion is that the serverless Inference API and dedicated Endpoints have separate reliability profiles and separate billing.
Where can I see Hugging Face incident history?
The official status page at status.huggingface.co keeps a running incident log with timestamps, affected components, and post-incident updates, which is the best source for past outages. The Hugging Face status account and the company's posts on X also surface major incidents in real time. For broader context on how Hugging Face reliability compares to other AI services over time, this page and TensorFeed's /status dashboard track the picture across providers.