Big Tech Moves

Hugging Face Hands Model Routing to OpenRouter

November 17, 2026 · By Editorial Team

Hugging Face Hands Model Routing to OpenRouter

Hugging Face has handed the routing of its language-model requests to OpenRouter. Announced on 16 November, the change replaces Hugging Face’s in-house Inference Providers router: every request to an open-weight language model now passes through OpenRouter, which picks a host from more than 80 providers based on live price, speed and reliability — and retries on another provider if the first one fails.

For developers, the practical pitch is continuity. Existing code and the same Hugging Face token keep working, and pricing is unchanged: users pay the provider’s list price with no markup on top. PRO users also get $2 of credits a month. The switch is the first step of a deeper partnership; linking a separate OpenRouter account and routing more task types through OpenRouter are planned next.

There is one hard break. Bring-your-own provider keys no longer work for language-model requests, which are now billed to the user’s Hugging Face account or organisation. Developers who previously paid a provider directly will now be billed through Hugging Face instead. Image, video and audio generation are untouched and keep working through the existing direct path, with custom keys still supported there — so image tools such as Midjourney are unaffected by the change.

The move lands as open-weight models dominate OpenRouter’s traffic: their share rose from 26.8% of tokens served in October 2025 to 71.7% by the end of September 2026. OpenRouter now serves more than 800 trillion tokens a month to over 10 million users. For anyone building on open models, the Hub just became a simpler front door — with fewer single points of failure behind it. It also strengthens the case for open-weight chat models such as DeepSeek, which developers can now reach through the same routed setup.

Source: Hugging Face Blog

Related stories