Hugging Face adds Public AI as a free, nonprofit serverless inference provider
Curated by the Inblix editorial team
Hugging Face just made it significantly easier to run models from public-interest institutions without setting up your own infrastructure. Public AI, a nonprofit inference utility, is now a supported Inference Provider on the Hub, meaning you can access models from groups like the Swiss AI Initiative and AI Singapore directly through serverless API calls.
The integration is baked into Hugging Face’s Python and JavaScript SDKs. You don’t need a separate Public AI API key to start—you can just use your standard Hugging Face token for automatic routing. A quick example shows calling Swiss AI’s Apertus-70B model with only a few lines of code, swapping the provider parameter to “publicai.” Behind the scenes, Public AI runs a distributed setup: a vLLM backend with OpenAI-compatible APIs, deployed across GPU clusters donated by national and industry partners. A global load balancer routes your query transparently, so you don’t have to think about whose hardware is actually doing the work.
Now, about the price tag. The announcement says usage is free at the time of writing, but that’s not the whole story. Free Hugging Face users get a tiny $0.10 monthly credit for inference. PRO subscribers get $2.00. That’s your buffer. If you burn through it, you’ll need to pay standard provider rates or use a direct Public AI key and settle up on your Public AI account. For now, the free credits are funded by donated GPU time and advertising subsidies. The long-term plan hopes to anchor stability on state and institutional cash—a model that feels refreshingly non-predatory but also precarious if grants dry up.
I’m genuinely bullish on this. It’s not just another API endpoint. It’s a deliberate counterweight to the commercial-only model access that’s become the default. Making sovereign, publicly funded models as trivial to call as GPT-4o lowers the barrier for researchers and builders who’d rather not feed the hyperscaler beast. The question is whether donated compute can keep up if this gets popular.”Free” is a great marketing word. Sustained, reliable free access is a whole different engineering challenge.
💡 Key Takeaways
- Public AI lets developers call sovereign models like Swiss AI’s Apertus-70B directly through Hugging Face’s standard SDKs by simply setting the provider to 'publicai'.
- The 'free' inference is actually capped by monthly credits: $0.10 for free users and $2.00 for PRO subscribers, after which standard provider rates apply.
- The entire backend relies on donated GPU time and vLLM-powered servers, making long-term reliability dependent on continued institutional funding rather than a commercial payment model.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.