Hugging Face Moves Language Model Routing to OpenRouter, Custom Provider Keys Stop Working
Summary
Hugging Face moved its Inference Providers routing for language models to OpenRouter, which picks among 80-plus providers by live price, speed and reliability and retries failed requests elsewhere. Hugging Face says billing is unchanged with no markup, and PRO users keep $2 in monthly credits. Custom provider API keys no longer work for language models.
Key Points
- OpenRouter reports its new routing cuts capacity errors 46% and improves time to first token 16.9%.
- Users pick providers with model id suffixes such as :fastest, :cheapest or a named provider like :novita.
- Image, video and audio generation keep running through Hugging Face directly, outside OpenRouter, with LoRA support intact.