Hugging Face adds Hyperbolic, Nebius, and Novita to its serverless inference army

Hugging Face just dropped a trio of new serverless inference providers into its platform: Hyperbolic, Nebius AI Studio, and Novita. They join an already crowded roster that includes Together AI, Sambanova, Replicate, fal, and Fireworks.ai. For developers, this isn’t just about having more logos on a settings page — it’s about direct, code-level access to specific models that were previously harder to reach, most notably DeepSeek-R1 and Black Forest Labs’ FLUX.1 image generation family.

The integration follows a clean pattern Hugging Face has been refining for months: bring your own API key from any provider and calls go straight to that provider’s infrastructure, bypassing Hugging Face’s routing layer. If you’d rather not juggle multiple provider accounts, just authenticate with your Hugging Face token and the platform handles the routing — with billing flowing through your HF account at the exact pass-through rate the provider charges. No markup. At least for now, though the blog post floats the possibility of future revenue-sharing deals.

From a code standpoint, switching providers is almost boringly simple. The Python and JavaScript SDKs expose a provider parameter that takes a string like “hyperbolic” or “nebius.” Change that string, and your request silently migrates to a different backend without touching model IDs or message formatting. The example in the announcement shows a one-line diff swapping Nebius for Hyperbolic when generating images with FLUX.1-schnell — the kind of developer experience that makes you wonder why API portability ever felt hard.

What’s genuinely useful here is the preference ordering. Users can rank providers in their account settings, and those preferences cascade into the model page widgets and code snippets. If Hyperbolic is your top pick for DeepSeek-R1 because of latency or pricing, the UI respects that. PRO subscribers get $2 in monthly inference credits they can burn across any provider, which feels modest but at least acknowledges that people want to experiment across backends without opening five separate billing relationships. The free tier remains, with a small quota for signed-in users, but the message is clear: serious usage should go PRO.