Hugging Face is replacing its own language-model router with OpenRouter, making open-weight models on the Hugging Face Hub available through a shared API and Hugging Face token. OpenRouter routes each request among more than 80 providers using live price, speed, and reliability data, and retries with another provider when one fails. Users can keep their existing code and use OpenAI Chat Completions or Responses, or Anthropic Messages APIs. They can also select fastest or cheapest routing, or pin a model to a specific provider. OpenRouter says its updated routing reduced capacity errors by 46%, improved time to first token by 16.9%, and increased throughput by 16.3%.
The partnership responds to the rapid growth and lower cost of open-weight models: their share of OpenRouter’s served tokens rose from 26.8% in October 2025 to 71.7% in September 2026, while average token prices were roughly four times lower than for closed models. The strongest open-weight models still score below top proprietary models on demanding benchmarks, but can cost substantially less per task. Hugging Face says pricing will not change and it adds no markup; users can inspect provider performance, spending, and routing settings, and block providers. A key change is that language-model requests are billed through Hugging Face, so users can no longer supply their own provider API keys. Image, video, and audio inference remain on the existing system, and connecting OpenRouter accounts and routing additional tasks are planned next.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.