Introducing Three New Serverless Inference Providers: Hyperbolic, Nebius AI Studio, and Novita 🔥

Hugging Face keeps stacking up inference options, and this week it added three more names to the list: Hyperbolic, Nebius AI Studio, and Novita. They join Together AI, Sambanova, Replicate, fal, and Fireworks.ai, which means the Hub now has eight serverless inference providers plugged directly into model pages and both client SDKs.

The practical upshot is access to models that weren't easily runnable before, like DeepSeek-R1 for text and Flux.1 for image generation. You pick a model on the Hub, and if one of these providers supports it, you can run it right there through a widget or a couple lines of code in Python or JavaScript, no separate account setup required unless you want one.

Users get two ways to pay. Bring your own API key from Hyperbolic, Nebius, or Novita and you're billed directly by that provider, same rates as going straight to them. Or route through Hugging Face with your HF token, and the charges land on your HF account instead, at the same price the provider charges, no markup added, at least for now. Hugging Face says revenue-sharing deals with providers could come later, but for the moment it's just acting as a pass-through.

Switching providers in code is almost trivially easy, just change one string, from provider="nebius" to provider="hyperbolic", and everything else in your script stays put. That kind of interchangeability is really the point here: Hugging Face wants to be the neutral layer where you shop between compute providers without rewriting your app each time.

PRO subscribers get $2 in monthly inference credits usable across any of these providers, and free users still get a small quota. Support for Hyperbolic lands officially in huggingface_hub v0.29.0, though it's already usable if you install from source now.