Hugging Face routes LLM inference through OpenRouter with automatic failover
·2026.11.17 12:16
Key point
Requests now route through OpenRouter for live price and speed optimization, with automatic retries on provider failure.
Details
- Hugging Face switched language model inference routing to OpenRouter on November 16, 2026, replacing its own router.
- OpenRouter selects providers based on live price, speed, and reliability, automatically retrying on a different provider if the first fails.
- Custom provider API keys no longer work for language model requests, which are now billed directly to the Hugging Face account without markup.
- Open-weight models accounted for 71.7% of tokens on OpenRouter by the end of September 2026, up from 26.8% in October 2025, and cost about four times less than closed models.
- Image, video, and audio generation remain on Hugging Face's existing Inference Providers path and do not use OpenRouter.
- Users can still pin specific providers or policies using suffixes like :fastest or :cheapest on the model ID.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.