
Spatial Intelligence in the Physical World
Today’s AI can see, but doesn’t comprehend. It recognizes objects but misses the critical spatial relationship that humans understand intuitively.
Ironsite is hosting a 36-hour hackathon to solve this problem. We’re looking for builders that can bridge the gap between seeing and understanding.
Pinpoint a key spatial task where current models fail. Find a visual task that is trivial for humans, but AI models completely fail at. Use existing models like Gemini 2.5 Pro, Claude Opus, or GPT5 to give them an image or video and ask them a question that reveals their lack of spatial intelligence. We will provide API credits for you to use.
Create an innovative solution to solve the problem. Utilize prompt engineering, simple fine-tuning, inference time compute strategies, or any other clever techniques you can think of to improve how well the models can perform on that task. Be detailed and scientifically minded.
Demo your technique on a real-world problem. Your idea doesn’t have to solve the problem perfectly, but rather show how these models can be augmented to increase their spatial intelligence, even for a narrow use case or task. It’s even better if the task you solve could have real world impact.

$17,500 (10,000 first prize, 5,000 second, 2,500 third)

November 14th (6 PM Kick-off) – November 16th (12PM Judging)

Teams of 3-5. Don’t have one? We’ll help you find your crew.

This is your chance to work on a frontier problem, get noticed, and win big.




A