Hugging Face Inference

Hugging Face Inference is the inference layer built into the Hugging Face Hub, covering two products: Inference Providers, which routes requests across 20+ third-party GPU backends including Groq, Cerebras, Together, and Replicate through a single API, and Inference Endpoints, which deploys any Hub model to a dedicated private HTTPS endpoint on AWS, GCP, or Azure in minutes. Founded in 2016 by Clément Delangue, Julien Chaumond, and Thomas Wolf, Hugging Face connects inference directly to the largest open-source model repository in ML, with 500,000+ models available without switching platforms.

Key Capabilities:

• Inference Providers multi-provider routing with automatic, cheapest, and preferred policies

• Access to 20+ named provider backends from a single InferenceClient call

• OpenAI-compatible API for switching from proprietary to open-source models in two lines of code

• Inference Endpoints for one-click dedicated GPU deployment with auto-scaling and scale-to-zero

• Text Generation Inference (TGI) open-source engine powering Inference Endpoints

• Python SDK via huggingface_hub and JavaScript SDK via @huggingface/inference

• Local endpoint support for Ollama, vLLM, llama.cpp, and TGI running on your own hardware

• Coverage across text generation, image generation, embeddings, speech recognition, and classification

• Dual authentication: routed through Hugging Face billing or direct with your own provider keys

• Interactive Inference Playground for testing chat completion models before integration

• Free tier with rate limits; PRO tier at $9/month for higher throughput

See Hugging Face Inference pricing details →