Hugging Face Inference Providers

source

closed

access

managed-cloud-API-only(routes to third-party providers)

client

MIT(@huggingface/inference)

license

Proprietary

Hugging Face publishes its client libraries, and the JavaScript client is MIT-licensed, but the router and its billing run only as a Hugging Face service. The terms of service treat Inference Providers as a limited right of access and use, not software you can run yourself.

• https://huggingface.co/docs/inference-providers/index recorded 2026-10-13

  The docs describe a hosted service reached with a Hugging Face token at https://router.huggingface.co/v1, which selects among partner providers; the page links no repository for the router, only github.com/huggingface and the hub-docs repository.

• https://huggingface.co/terms-of-service recorded 2026-10-13

  The terms say Services "may include limited licenses or subscriptions to access or use certain offerings ... including ... the Inference Providers" and that purchases "mean a limited right to access and use a Service (not a transfer or any ownership right, title, or interest)".

• https://registry.npmjs.org/@huggingface/inference/latest recorded 2026-10-13

  The npm registry metadata for @huggingface/inference 4.13.32 gives license "MIT", repository huggingface/huggingface.js and the description "Typescript client for the Hugging Face Inference Providers and Inference Endpoints".

Hugging Face publishes no request, token or user count for Inference Providers. Its JavaScript client also serves Inference Endpoints and its Python client the whole Hub, so their downloads cannot be credited to the router alone.

• https://huggingface.co/docs/inference-providers/pricing recorded 2026-10-13

  The pricing page, read for a usage disclosure, describes pay-as-you-go billing, monthly credits for PRO, Team and Enterprise accounts and "no markup from Hugging Face"; it gives no request, token or user figure.

• https://registry.npmjs.org/@huggingface/inference/latest recorded 2026-10-13

  The npm package describes itself as the "Typescript client for the Hugging Face Inference Providers and Inference Endpoints", so its installs span two services.

Inference Providers runs no models itself, but it reaches the fastest specialist hosts, Fireworks among them, and can choose between them on each request by speed or price. What it can do is roughly what those providers can, so it sits alongside Fireworks rather than above the services it depends on.

• https://huggingface.co/docs/inference-providers/index recorded 2026-10-13

  The provider table lists Baseten, Cerebras, Cohere, DeepInfra, Fal AI, Featherless AI, Fireworks, Groq, HF Inference, Novita, Nscale, OVHcloud AI Endpoints, Public AI, Replicate, Scaleway, Together, WaveSpeedAI and Z.ai across chat completion (LLM, VLM), feature extraction, text to image, text to video and speech to text; "By default, our system automatically selects the fastest available provider", with ":cheapest" or a provider name such as "openai/gpt-oss-120b:groq" as alternatives, and the page calls it a "Drop-in replacement for the OpenAI chat completions API".