Hugging Face

What Plutus imports

• Inference Endpoint compute cost by model, region, and instance type (estimated)

Connecting

Method: User Access Token

1. In Hugging Face, go to Settings > Access Tokens.
2. Click New token, set the type to Read, and give it a name.
3. Copy the token.
4. In Plutus, paste it into Access Token.
5. Optionally enter your Organisation Name — required if you want to pull organisation-level billing data rather than just your personal account’s.
6. Optionally set Instance Rate Overrides if you’re on negotiated or Enterprise pricing: comma-separated instance=rate pairs in USD per hour, for example nvidia-a100=2.10, aws/nvidia-l40s/x4=7.50. The instance key is the label shown in the Instance Type breakdown (vendor/type/size), or any shorter prefix of it — nvidia-a100 matches every A100 size.
7. Optionally set a Fallback Rate per Hour for instance types Hugging Face hasn’t published a rate for and that no override matches (typically newly released hardware). This rate is scaled by instance size, so a fallback of 1.00 prices an x4 instance at $4.00/hr.
8. Save the connection.

Plutus lists your Inference Endpoints and prices each one against Hugging Face’s published hourly rate for its exact instance — vendor, instance type, and size — multiplied by its current replica count.

Notes

HuggingFace has no billing API of any kind, so cost is estimated, not invoiced: each Inference Endpoint is priced against HuggingFace’s published per-instance hourly rate card and its current replica count. That rate card is a snapshot held in Plutus, so it can lag a price change, and replica counts are read as they are now rather than as they were on each past day — an autoscaling endpoint reads as flat. Pro/Enterprise subscription fees, serverless Inference Providers, Jobs and Spaces hardware are not visible to any API and are not included.