Hugging Face routes language-model requests through OpenRouter across more than 80 providers
Developers keep their existing code and token, and can switch off individual providers, while image, video and audio generation stay on the current path.
Hugging Face is moving its Inference Providers language-model traffic onto OpenRouter's routing layer, so requests now go to whichever of more than 80 providers currently offers the best balance of price, speed and reliability for each call. A failed request is retried on another provider within the same call, and developers keep their existing code and Hugging Face token; open-weight models that a provider serves on the Hub can be reached this way. OpenRouter's figures put open-weight models at 71.7% of tokens served by late September 2026, up from 26.8% in October 2025, and Hugging Face says open-weight models average about four times cheaper per token than closed ones. Billing passes provider prices through with no Hugging Face markup, and users can disable individual providers in their settings. Image, video and audio generation stay on the existing Inference Providers path, and this is described as the first step of the partnership, with connecting OpenRouter accounts and routing more task types planned next.
Why it matters
The change makes provider choice a routing decision rather than a fixed setting, so the price and reliability of open-weight models matter more to who serves a given request. Open-weight models accounted for 71.7% of tokens served on OpenRouter by late September 2026, up from 26.8% in October 2025, so this shift affects a growing share of usage. Because billing passes provider prices through without markup, the savings and the choice of provider land directly on the user's Hugging Face bill.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.
Sources · 1
- Hugging Face Blog ↗ Primary source