Product Launch3 min read

Hugging Face switches its inference router to OpenRouter

Language-model calls made through Hugging Face Inference Providers now route through OpenRouter, which picks a provider per request by price, speed and reliability. Hugging Face says it adds no markup, but users can no longer bring their own provider API keys for LLMs.

AE
Aidose Editorial
Nov 17, 2026

Hugging Face announced on November 16 that it is replacing its own custom router for Inference Providers with OpenRouter's. Language-model requests made with a Hugging Face token now pass through OpenRouter, which selects a provider for each call based on price, speed and reliability and retries on another provider if one fails.

The API surface does not change, according to Hugging Face: every chat model supports OpenAI's Chat Completions and Responses APIs as well as Anthropic's Messages API, and the company has published setup guides for Claude Code, Codex, OpenCode, Pi and Hermes Agent. Developers can steer routing by appending :fastest, :cheapest or a provider name such as :novita to a model ID, and can send a stable X-HF-Session-Id header per conversation to keep requests on the provider holding their prompt cache. Model pages gain Providers and Metrics tabs showing each provider's price, speed and uptime.

Users pay the provider's price on their Hugging Face bill, and "Hugging Face doesn't add a markup on top of it." PRO users get $2 of monthly credits, and Team and Enterprise organizations get $2 per seat. The change comes with one breaking change: for language models, users can no longer supply their own provider API keys, and billing must go through a Hugging Face account or organization. Providers can still be pinned with a suffix or switched off in settings. Image, video and audio models are unaffected and continue to run directly through Inference Providers.

The post cites OpenRouter figures of more than 800 trillion tokens a month, over 10 million users, 600+ models and 80+ providers, and says OpenRouter's rebuilt routing cut capacity errors by 46% and improved time to first token by 16.9%. It also says open-weight models accounted for 71.7% of OpenRouter tokens at the end of September 2026, up from 26.8% in October 2025, with average open-weight prices of $0.28 input and $1.62 output per million tokens against $1.27 and $6.36 for closed models.

For teams already calling open-weight models through Hugging Face, the practical effect is automatic failover and price-based routing without code changes, at the cost of losing direct billing with individual providers for LLM traffic. Hugging Face says it plans to let users connect OpenRouter accounts and to route more task types through OpenRouter.

AE
Aidose Editorial
Nov 17, 2026 · 3 min read
Back to News