Hugging Face

1:31 Hugging Face is the website where most open AI models are published and downloaded: its Hub holds more than 2 million models, 500,000 datasets and 1 million apps (as of Nov 6, 2026), from Meta's Llama and Google's Gemma to DeepSeek, Qwen, Mistral and OpenAI's gpt-oss. Browsing, downloading and using free apps cost nothing; PRO costs $9 a month, Team $20 a user and Enterprise $50 a user. Beyond the Hub it sells hosted inference through Inference Providers, runs free GPU demos on Spaces, and offers the HuggingChat assistant.

Use Hugging Face to find and download open models (directly or through Ollama and LM Studio, which pull from it), to try a model in a browser before installing it, or to call open models through one API. For a single API that also covers closed models such as GPT, Claude and Gemini, OpenRouter is the closer fit; for everyday chat, a mainstream assistant such as ChatGPT or Claude is easier than HuggingChat.

Quick facts

The Hub: models and datasets

2:25 Each model on the Hub has a page with its model card (what it is, how it was trained, its license), the files, a download counter and often a live widget to try it. Model makers publish there first: Gemma 4, gpt-oss, Mistral Large 4 and Llama 4 all ship their weights on it. Some models, such as Llama, ask you to accept a license before downloading. Trending models on Nov 6, 2026 included Moonshot's Kimi K3.1 Pro and Lightricks' LTX-2.5 video model. Datasets work the same way for training and evaluation data.

Spaces and ZeroGPU

12:54 Spaces are small web apps, usually model demos, that anyone can run in a browser. Many run on ZeroGPU, shared Nvidia RTX Pro 6000 Blackwell GPUs that Hugging Face hands out per request. Using ZeroGPU Spaces is free with a daily GPU quota: 2 minutes without an account, 5 minutes with a free account and 40 minutes on PRO, after which PRO users pay $1 per 10 minutes. Free accounts older than 30 days can host 2 ZeroGPU Spaces; PRO can host 10. Paid Space hardware runs from $0.03 an hour for a CPU upgrade to $2.50 an hour for an A100.

Inference Providers

3:49 Inference Providers is one API, with an OpenAI-compatible address (router.huggingface.co/v1), that sends requests for open models to partner services: Cerebras, Groq, Together, Fireworks, Novita, Nscale, Replicate, fal, Scaleway, OVHcloud, Z.ai and others. It picks the fastest provider by default, or the cheapest if you add :cheapest to the model name. Hugging Face charges the providers' own rates with no markup. PRO users and Team and Enterprise seats get $2 of compute credits a month; free users buy credits to use it. It also works as the back end for coding agents such as Claude Code and Codex, per the same docs.

HuggingChat

6:37 HuggingChat is Hugging Face's free chat assistant for open models. On Nov 6, 2026 its page listed 128 models, an "Omni" mode that picks a model for each request, and a new "ML Intern" agent for machine learning work (papers, training, datasets and evaluations). It is a good way to try many open models side by side; it lacks the memory, apps and polish of ChatGPT, Gemini or Claude.

Plans and prices

Prices as of Nov 6, 2026. Storage beyond the free allowance costs $12 per TB a month for public and $18 for private repositories, with discounts from 50 TB.

Hugging Face vs the alternatives

• OpenRouter: one API for open and closed models alike, with public rankings of token use; pick it if you also need GPT, Claude or Gemini.
• Ollama and LM Studio: run the Hub's models on your own computer instead of in the cloud.
• Kaggle and GitHub: some makers (Meta, Google) also publish weights or code there, but the Hub is where most open models live.

Frequently asked questions

Is Hugging Face free? Yes. Browsing and downloading models and datasets on Hugging Face, and using free Spaces, cost nothing; PRO is $9 a month for more GPU quota, storage and $2 of monthly compute credits, as of Nov 6, 2026.

What is the Hugging Face Hub? The Hugging Face Hub is the site where AI makers publish open models, datasets and demo apps; it held more than 2 million models on Nov 6, 2026.

Is HuggingChat still available? Yes. HuggingChat at huggingface.co/chat listed 128 open models on Nov 6, 2026, with an Omni mode that picks the model for you.

How do I use a Hugging Face model? Download it with Hugging Face's tools or an app such as Ollama or LM Studio, try it in a Space in your browser, or call it through Hugging Face's Inference Providers API.

Does Hugging Face charge extra for Inference Providers? No. Hugging Face says it passes on each provider's rates with no markup; free users buy credits, and PRO users get $2 of credits a month.

Sources

• Hugging Face, read Nov 6, 2026
• Pricing, Hugging Face, read Nov 6, 2026
• Inference Providers, Hugging Face docs, read Nov 6, 2026
• Pricing and Billing, Hugging Face docs, read Nov 6, 2026
• Spaces ZeroGPU, Hugging Face docs, read Nov 6, 2026
• HuggingChat, Hugging Face, read Nov 6, 2026
• What is Hugging Face? - Models, Datasets & Spaces, Hugging Face, Mar 13, 2026 (video)