Language Models are Limited
Even the most advanced LLMs fail at the simplest tasks for kids.
Video Models Start to Reason
The real substrate of spatiotemporal intelligence has arrived.
“Navigate through the maze from green start to red flag end.”
56 authors · 32 universities · 5 continents
Language is a subtask of video reasoning.
Function tool calling leads to language model agents. We wonder what is the model context protocol for video models.
Active Products
Monetization over Creative and Data first.
We play the Anthropic game, where monetization over chatbot is a day-one strategy.
Potato Studio
Draw on images to generate videos.
Data Fountain
Generate video reasoning datasets on demand.
50+
Artists signed up
Potato Studio
7+
Enterprises in conversations
Data Platform
Vertical Domains Are When The Actual Game Starts
Our continued foundational research guided by deep understanding of intelligence will find us the Claude Code moment.
Manufacturing & Factories
Health & Medicine
Sports Analytics
Weather & Climate
Founder & CEO
Hokin Deng
Hokin built VBVR from scratch — the benchmark, the data pipeline, the model, the evaluation framework, and the research community around it, now leading 55 researchers from 32 universities across 5 continents — no lab, no grants, just the work.
Unfair Advantage
100 university Discord channels. 80 WhatsApp groups. One afternoon to mobilize 500 contributors. We call it the Boba Tea strategy — and no data platform on earth can match our cost or speed.