Hugging Face routes language-model inference through OpenRouter
First merged summary.
Hugging Face now routes language-model inference through OpenRouter, ending bring-your-own language-model keys while image, video, and audio stay direct.
Hugging Face now routes language-model requests on Inference Providers through OpenRouter, which chooses a provider by price, speed, and reliability and fails over in outages. Existing integrations keep working, and :fastest, :cheapest, or the X-HF-Session-Id header can steer routing. Bring-your-own keys for language models are ending; image, video, and audio generation still run directly. OpenRouter says capacity errors fell 46 percent, and Hugging Face says open models rose from 26.8 percent to 71.7 percent of top-model tokens. According to The Decoder, Nvidia’s planned $12.9 billion purchase of Hugging Face is under DOJ review, and Stripe agreed in August to buy OpenRouter for more than $7 billion; neither deal has closed.
- Language-model requests now route through OpenRouter by price, speed, and reliability.
- Bring-your-own language-model keys are ending; image, video, and audio stay direct.
- Suffixes :fastest and :cheapest, or X-HF-Session-Id, steer provider choice.
- OpenRouter says capacity errors fell 46 percent with the new routing.
- Hugging Face says open models rose from 26.8 to 71.7 percent of top-model tokens.
- Neither Nvidia’s reported $12.9 billion deal nor Stripe’s OpenRouter purchase has closed.
Timeline
dated events in the reporting · oldest first- Nov 16, 2026Hugging Face announces partnership to ease open-weight model use · Hugging Face Blog
- Nov 18, 2026Decoder details OpenRouter routing and ending language-model keys · The Decoder
What each outlet adds
beyond the shared facts- Hugging Face BlogStates the partnership goal without routing or product mechanics.
- The DecoderAdds routing rules, key changes, usage stats, and unclosed acquisition reports.
- Products
- Inference ProvidersOpenRouter
- AI models
- Kimi K3.1 ProClaude Fable 5.2
Coverage
oldest first · one row per article · UTCHugging Face and OpenRouter partner to ease running open-weight models
The Hugging Face blog post announces the tie-up, but the available text gives no product or launch details.
Hugging Face routes language model requests through OpenRouter
Bring-your-own keys for language models no longer work, while image and video keys remain.