Oriel — AI & technology worth noticing, every day.

Hugging Face and OpenRouter join forces to make open-weight models easy to run

Web page · https://huggingface.co/blog/openrouter

First seen by Oriel on .

AI summary

Hugging Face is switching language-model Inference Providers to OpenRouter, which dynamically routes open-weight model requests across providers using live price, speed, and reliability measurements.

  • The same Hugging Face token and API support OpenAI Chat Completions, Responses, and Anthropic Messages APIs across chat models and providers.
  • OpenRouter serves over 800 trillion tokens monthly across 600+ models and 80+ providers, with routing that reduced capacity errors by 46%.
  • Open-weight models averaged $0.28 per million input tokens and $1.62 per million output tokens, roughly one-quarter the cost of closed models.
  • Requests can specify routing policies such as :fastest, :cheapest, or a particular provider, while stable X-HF-Session-Id values preserve prompt-cache locality.
  • Language-model requests can no longer use custom provider keys, but image, video, and audio generation continue using Hugging Face Inference Providers directly.

#open-weight models #openrouter #inference routing #hugging face

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
Hugging Face Bloghttps://huggingface.co/blog/openrouterhttps://huggingface.co/blog/openrouter Read discussion