Hugging Face integrates OpenRouter to unify access to open-weight AI models

What happened
Hugging Face has integrated OpenRouter to handle all inference requests for open-weight AI models on its platform. This change allows developers to access any open-weight model through a single API and token, with OpenRouter automatically selecting the best provider for each request based on price, speed, and reliability. The integration was implemented in October 2026, and users' code and pricing remain unchanged. Image, video, and audio generation continue to be served directly by Hugging Face and are not affected by this change.
Key details
- OpenRouter routes requests across more than 80 providers and 600+ models.
- Open-weight models made up 71.7% of tokens served on OpenRouter by September 2026, up from 26.8% in October 2025.
- The most used model on OpenRouter is DeepSeek V4.1 Flash 1017, with 59.2 trillion tokens served.
Why it matters
Developers using Hugging Face can now access a wider range of open-weight models more reliably and at lower cost, without changing their existing code or how they are billed. Organizations and individual users benefit from improved failover, caching, and the ability to select or exclude specific providers, giving them more control over performance and spending. The integration also means users can no longer use their own API keys for language model providers, centralizing billing and provider selection through Hugging Face.

