Gradio just shipped its 3.0 release, and it’s not a minor version bump — it’s a ground-up rebuild of the library that has powered more than 600,000 machine learning demos since 2019. The frontend now runs on Svelte, which the team says delivers much smaller payloads and faster page loads. That matters more than it sounds: demos embedded in blog posts or Hugging Face Spaces have been getting sluggish, and lighter builds mean more people actually stick around to try a model.
The visual overhaul also brings a cleaner design that fits more naturally into embedded contexts, plus revamped components like a drag-and-drop Dataframe and a new Gallery component for image-heavy models. A TabbedInterface class now lets developers group related demos under one roof instead of linking out to separate apps.
But the headline feature is Blocks, a new low-level API that gives Python developers fine-grained control over layout and data flow. Standard Gradio interfaces lock you into inputs on the left and outputs on the right. Blocks breaks that constraint: you can chain outputs from one model into the inputs of another, toggle component visibility based on user input, or wire a text box to update on every keystroke rather than waiting for a submit button. The team is clearly betting that flexibility is what keeps sophisticated ML builders on the platform — and hoping it pulls in developers who previously outgrew Gradio’s opinionated structure.
To push adoption, Hugging Face is running a Blocks Party competition through the end of May, offering merch and prizes for demos built on the new API. The timing is smart. With so many open-source models shipping weekly, the bottleneck for adoption is increasingly the demo layer — someone has to make a state-of-the-art model feel approachable in a browser. Gradio 3.0 is a direct shot at making that easier, and the competition is a cheap way to flood the ecosystem with examples before the API even has full documentation traction.
