Hugging Face switches model routing to OpenRouter
Hugging Face is moving Inference Providers routing to OpenRouter, adding live provider selection and failover while changing how language-model requests are billed.
Hugging Face is switching its Inference Providers routing service to OpenRouter, bringing provider selection, failover and pricing information into the same path used to serve open-weight models on the Hub. The change was announced on 16 November and applies to language-model requests, while the companies say existing code can continue using the same APIs. 1
Under the new arrangement, OpenRouter will choose a provider for each request using live measurements of price, speed and reliability. It can retry a request with another provider if the first one fails, according to Hugging Face. The service supports OpenAI’s Chat Completions and Responses APIs, as well as Anthropic’s Messages API, for chat models regardless of which provider runs the selected model. 1
The practical pitch is convenience. Developers can use a Hugging Face token to call open-weight models that are served by a provider, rather than building separate routing logic around each provider. Hugging Face lists integrations for Claude Code, Codex, OpenCode, Pi and Hermes Agent, alongside its own inference documentation. 1
There is a significant billing change. Language-model requests are now charged to the Hugging Face account or organisation, and users can no longer use their own provider API keys for those models through Inference Providers. Teams will therefore need to understand how Hugging Face’s account-level billing and OpenRouter’s provider selection interact before moving production workloads. 1
Why OpenRouter is taking over the routing
Hugging Face says open-weight models made up 71.7 per cent of tokens served by the top 50 models on OpenRouter in September 2026, up from 26.8 per cent in October 2025. Those figures come from the companies’ own analysis and are not an independent market measurement, but they explain why the partnership is being positioned as infrastructure for the growing open-model ecosystem. 1
Our opinion
This looks like a sensible infrastructure trade if OpenRouter’s live provider measurements really translate into fewer failed or overpriced requests. The catch is control: developers gain a simpler interface but lose the ability to supply their own provider keys for these language-model calls. That makes billing, privacy and provider-selection policies worth checking before treating the change as a frictionless upgrade.