The Hugging Face Inference API lets you run models hosted on the Hub via HTTP — no GPU, no setup, no infrastructure.

Free and Serverless Tiers

• Free Inference API: Rate-limited access to most Hub models
• Serverless Inference API: Pay-per-request, higher limits
• Dedicated Endpoints: Your own private, always-on endpoint

Python Client

from huggingface_hub import InferenceClient

client = InferenceClient(token="hf_...")
response = client.text_generation(
    "Explain quantum computing in simple terms",
    model="meta-llama/Meta-Llama-3-8B-Instruct",
    max_new_tokens=200
)
print(response)

Reference:

Inference API documentation

https://huggingface.co/docs/api-inference