ToolNavs Find Useful AI Tools
Submit Sign in
Back to AI information
Hugging Face Moves Inference to OpenRouter: One Call to an Open Model, the Best Provider Picked Automatically

Hugging Face Moves Inference to OpenRouter: One Call to an Open Model, the Best Provider Picked Automatically

AI information • Admin • • 14 views

On November 16, 2026, Hugging Face announced on its official blog that its inference service, Inference Providers, is switching from its in-house router to OpenRouter with immediate effect: developers keep using the same Hugging Face token and the same API to call open-weight models on the Hub, while each request is now routed by OpenRouter, which picks a provider based on live price, speed, and reliability — and automatically switches to another provider if one fails. No code changes are needed.

Same Way of Calling, a New Hand Picking the Provider

The new router supports OpenAI's Chat Completions and Responses APIs as well as Anthropic's Messages API, and coding agents such as Claude Code, Codex, and OpenCode can connect directly. Model pages gain two new tabs: Providers lists the providers OpenRouter can route a model to, with each one's price, speed, and availability; Metrics shows the best price, throughput, time to first token, and context length. Providers you don't want can be switched off in settings, and conversations stay with the provider holding the prompt cache.

Open Models Already Account for 70% of OpenRouter Traffic

The numbers Hugging Face cites explain the confidence behind the move: open-weight models' share of tokens served by OpenRouter rose from 26.8% in October 2025 to 71.7% at the end of September 2026, peaking at 77.5% in July. Over the same period, weekly open-model volume grew more than 60-fold, against under 10-fold for closed models. This week, the most-used model on OpenRouter is the open DeepSeek V4.1 Flash 1017, at 59.2 trillion tokens. On price, open models average $0.28 and $1.62 per million input and output tokens, versus $1.27 and $6.36 for closed models — roughly four times cheaper. The trade-off is capability: on the Artificial Analysis Intelligence Index, the best open model, Kimi K3.1 Pro, scores 46 against 65 for the best closed model, Claude Fable 5.2; but a single task costs just $3.68, less than a seventh of the other side's figure.

There is one breaking change: language-model requests are now billed through OpenRouter to your Hugging Face account, and developers can no longer attach their own provider API key to these models. Image, video, and audio generation are unaffected. Pricing itself does not change — Hugging Face adds no markup, and PRO users still get $2 of credits a month. The switch comes after Stripe completed its acquisition of OpenRouter. Next, Hugging Face plans to support connecting an OpenRouter account directly and to bring more tasks into this routing. For developers, choosing a model is shifting from 'picking a provider' to 'picking a model,' and the contest comes back to price, speed, and reliability.

Recommended Tools

More