Hugging Face just dropped a free, open-source rival to OpenAI's GPTs

Building a custom AI agent used to come with an OpenAI tax: at least $20 a month for ChatGPT Plus, or $25 per seat on a Team plan.

Now, open-source hub Hugging Face is kicking open the door with Hugging Chat Assistants, a free maker that lets anyone build and publish custom AI chatbots without spending a dime.

Two clicks to build, zero dollars to host

Hugging Face tech lead and LLMs director Phillip Schmid announced the launch on X, openly pitching it as an open alternative to OpenAI's custom GPTs. According to Schmid, users can spin up an assistant in just two clicks.

The setup looks immediately familiar. Hugging Face even rolled out a public aggregator that looks nearly identical to OpenAI's GPT Store, complete with baseball card-style rectangular boxes and circular avatar icons.

But the biggest differentiator sits under the hood.

While OpenAI locks your custom bot to proprietary models like GPT-4 or GPT-4 Turbo, Hugging Face lets you pick your backend engine. You can run your assistant on Meta's Llama 2, Mistral's Mixtral, or whichever open model best fits your workflow, with Hugging Face footing the inference bill.

What's missing from the open sandbox

The open-source crowd was quick to celebrate. Mathieu Trachino, founder of enterprise AI firm GenDojo.ai, took to X to call the release better than GPTs in terms of model flexibility and sheer affordability.

Still, it arrives with noticeable trade-offs.

Unlike OpenAI's mature builder, Hugging Chat Assistants currently lack web search and retrieval-augmented generation (RAG) for querying custom documents. You also can't auto-generate bot logos with an image model like DALL-E 3, so you have to bring your own image. Reports note that RAG and web browsing are at least on the roadmap.

Why it matters

We think Hugging Face just wiped out the entry fee for building functional AI personas.

For everyday users and prompt engineers, paying $20 a month just to share a prompt wrapper or a study tutor felt steep. Hugging Chat Assistants turns that into a zero-cost playground, while giving builders real leverage to switch model brains whenever open weights improve.

For the industry, it is another clear signal that OpenAI's distribution lead is being squeezed from both sides. Just days after Mistral's leaked Miqu model showed open weights closing the gap with GPT-4, Hugging Face proved that OpenAI's marketplace playbook can be cloned overnight.

That said, we see an obvious catch: free inference has limits. Without RAG and live browsing, these assistants are basically dressed-up system prompts for now. They are great fun to try today, but builders waiting for serious workflow automation will need to hold out for those roadmap updates.