Big news! OpenBMB teams up with Hugging Face & Gradio to launch the Build Small Hackathon, with $40,000 in prizes

Dear developer friends:

Organized by Hugging Face & Gradio, OpenBMB will serve as an Anchor Sponsor, joining OpenAI, NVIDIA, and other global AI communities and organizations to support the Build Small Hackathon.

This is a fully online global developer hackathon. Its theme is very direct, and very much worth rethinking for today's AI developers:

Build Small.

Use small models to build AI applications that are small and beautiful, small and useful, small and fun.

➤ Event website

🔗 https://huggingface.co/build-small-hackathon

Highlights

Build Small Hackathon is hosted by Gradio × Hugging Face, with OpenBMB, OpenAI, and NVIDIA as Anchor Sponsors. The total prize pool exceeds $40,000, including cash and in-kind awards; OpenBMB will provide $10,000 special category prizes for outstanding projects using MiniCPM series models.

Why Build Small?

Over the past year, the main theme of the AI world seems to have been:

Bigger models.

More parameters.

More complex systems.

Higher compute barriers.

But we increasingly hear another voice:

AI should not belong only to super-large models, super-large compute, and super-large companies.

The most vital AI applications do not necessarily come from the largest models. They can come from a small model that is smart enough, lightweight enough, and useful enough, running on your computer, local device, mobile phone, or even a simple Hugging Face Space.

That is exactly what draws us to Build Small Hackathon.

It does not encourage another giant, distant, hard-to-deploy AI system. It wants developers to return to that familiar creative state:

Start fast, validate fast, solve one concrete problem, or build something that genuinely makes people smile.

Small models do not mean small imagination. Small models can solve real problems, create new interaction experiences, and make AI playful, usable, and approachable again.

Two tracks: solve real problems, or create pure joy

This hackathon has two main tracks.

🏡 Track 1: Backyard AI

This track focuses on the real people and problems around you.

You can build an AI application that truly helps your parents, neighbors, classmates, the owner of a small shop on the street, or real users in your community.

For example:

Help parents understand phone screenshots, bills, and manuals;

Help a small shop owner organize menus, receipts, and product information;

Help students understand a photo of a homework problem;

Help ordinary people read complex tables, documents, and repair manuals;

Build a lightweight AI assistant that runs locally.

This track does not care how grand the concept is. It cares about:

Is the problem real, are the users real, and does the application actually run?

🍄 Track 2: An Adventure in Thousand Token Wood

This track encourages stranger, more playful, more imaginative things.

It could be a small AI game, an interactive story, a quirky but addictive toy, a fantasy character companion app, or any AI experience that makes people want to share it with friends.

The point is not "what AI does for you," but:

Would this thing exist at all without AI?

AI must be the core of the experience, not decoration stuck on the outside.

MiniCPM Special Awards: $10,000

MiniCPM is OpenBMB's series of efficient models, covering language models, multimodal understanding, omni-modal interaction, reasoning, and speech generation.

Its core characteristics:

Smaller model sizes;

Lower deployment barriers;

Better suited to local and on-device applications;

Support for multimodal understanding and interaction;

Suitable for developers building small but complete AI applications quickly.

This aligns closely with the spirit of Build Small Hackathon:

Not the biggest, but genuinely usable.

Not just demonstrating model capabilities, but delivering real experiences.

Not just stacking parameters, but bringing AI to everyone.

As an Anchor Sponsor of Build Small Hackathon, OpenBMB will provide $10,000 MiniCPM Special Awards for outstanding projects in the two tracks that use MiniCPM series models. We want to see more developers use MiniCPM models and explore what small models can do in real applications.

You can use:

MiniCPM-V 4.6: image understanding, OCR, document understanding, multimodal applications;

MiniCPM-o 4.5: omni-modal interaction, speech + vision + text experiences;

MiniCPM-V 4.5: multimodal understanding, visual reasoning, image/video applications;

MiniCPM5-1B: lightweight text generation, local applications, on-device AI assistants;

MiniCPM4.1-8B: text reasoning, problem solving, local inference applications;

VoxCPM2: speech generation, TTS, character voices, and creative voice experiences;

Other eligible MiniCPM / OpenBMB series models.

➤ MiniCPM series models

🔗 https://huggingface.co/openbmb/collections

How to participate

The hackathon process is clear:

Join the official Build Small Hackathon Hugging Face Org;

Build a Gradio App using a small model;

Host the app as a Hugging Face Space;

Submit the Space link, a short demo video, and a social media post;

If you used a MiniCPM model, clearly state in the README which model you used, how you used it, and the relevant evidence.

The event page also states clearly that joining the official Hugging Face Org is required to participate.

➤ Registration

🔗 https://huggingface.co/build-small-hackathon

Final thoughts

The lightness of AI creation has never disappeared; it has always been there.

In developers' weekend projects, in a student's first demo, in a tool that helps parents solve a small problem, in a little AI toy that makes a friend smile.

It is just that for a while, we too easily imagined AI as bigger models, stronger compute, and more complex systems.

Build Small reminds us:

AI does not always have to be a towering giant.

It can also be a small tool, a small assistant, a small game, an app that helps parents solve a problem, or a demo that makes a friend smile.

Developers, students, and creators interested in small models, multimodality, on-device AI, Gradio, and Hugging Face Space are all welcome to join us.

For any questions about the event, scan the QR code to join the Feishu group and post for help:

Use MiniCPM. Build Small.

We'll see you at Build Small Hackathon.

➤ Event website

🔗 https://huggingface.co/build-small-hackathon

➤ MiniCPM models

🔗 https://huggingface.co/openbmb/collections