Hugging Face

What's good about Hugging Face

• Deploy any compatible Hub model to dedicated GPUs without managing infrastructure
• Choice of cloud and region per endpoint
• Inference Providers adds no markup over partner rates
• Monthly inference credits on free and PRO accounts

Hugging Face GPU prices

8 GPU models across 36 configurations. Rank compares Hugging Face's cheapest on-demand listing for each GPU with every provider's cheapest.

What a run costs on Hugging Face

24 hours at today's cheapest listing for each GPU, at the size it is listed.

• 1× H100 SXM for 24 hours

  gcp, us-east4 at $10.00/GPU-hr

  $240.00

  $161.28 over the market median rate

• 1× H200 SXM for 24 hours

  aws, us-west-2 at $5.00/GPU-hr

  $120.00

  $13.56 over the market median rate

• 1× A100 SXM 80GB for 24 hours

  aws, us-east-1 at $2.50/GPU-hr

  $60.00

  $21.72 over the market median rate

The market median is every provider's cheapest on-demand price for that GPU.

Billing & terms

Inference Endpoints

Per hour of GPU instance time, billed by the minute

Inference Providers

Pass-through of the partner provider's per-token rate

Enterprise

Team and Enterprise Hub plans with managed billing

Support

Documentation, community forums, and paid support on Enterprise plans

Watch out for

• Endpoint GPU rates are above raw cloud rental for the same card
• GPU availability differs by cloud and region
• Inference Providers rates are the partners' rates, so costs vary with the provider a request is routed to

Features

Inference Endpoints

Dedicated deployments of Hub models on NVIDIA GPUs (T4 to B200 and RTX PRO 6000) on AWS or Google Cloud, with a choice of region and 1 to 8 GPUs; AWS Inferentia2 and Google TPU v5e instances are also offered

Inference Providers

One API and token for models served by partner providers such as Together, Fireworks, Groq, Cerebras, Novita and DeepInfra, billed at each provider's rate with no Hugging Face markup

Model Hub

Hosting for open models, datasets, and Spaces demo apps

Autoscaling

Endpoints scale replicas with traffic and can scale to zero

Getting started

1. Step 1

   Create an account

   Sign up at huggingface.co and add a payment method

2. Step 2

   Deploy an endpoint

   Pick a model, cloud, region, and GPU instance in the Inference Endpoints console

3. Step 3

   Or use Inference Providers

   Call a model through the router with a Hugging Face access token