Drawing Arena
You draw. AIs race to guess in real time.
Start new thread

Drawing Arena - You draw. AIs race to guess in real time.

by•
Drawing Arena turns a 60-second sketch into a live AI race. You get a secret word, draw it, and independent AI models try to identify your picture every five seconds—without seeing each other’s answers. Mystery matchups keep model identities hidden until the reveal, and every finished race can become a replay challenge for friends. It’s a game first, but also a fun way to see how differently vision models interpret the same drawing.

Add a comment

Replies

Best
Maker
📌
I built Drawing Arena around a simple question: what happens when you give multiple AIs the exact same messy human drawing and make them race to understand it? Each model sees only the canvas, not the secret word or another model’s guesses, and gets a fresh chance every five seconds. The result is part game, part live vision-model experiment, and often surprisingly funny. I wanted the experience to feel understandable even to someone who doesn’t care about benchmarks: draw something, watch the AIs think, see who gets it first, then challenge a friend with the same replay. This is an early launch, so I’d especially love feedback on the game loop, model behavior, and what you’d want to see next.
OpenAI
GPT-6 Astra Challenge
What became possible in your product with Astra that was not practical before?
Maker
Before Astra, Drawing Arena was easy to imagine but much harder to turn into a reliable live product quickly. Astra made it practical to iterate across the whole stack at once: real-time drawing, isolated AI model lanes, guess timing, replay sharing, authentication, deployment, and adversarial testing. The biggest unlock was being able to reason across all of those interacting systems together—finding edge cases like duplicate guesses, tracing failures from browser to backend, and shipping fixes with verification instead of manual trial-and-error. That let us turn the idea into a working multi-model game, not just a prototype: a human draws once, multiple AIs independently interpret the same evolving image, and the result becomes something you can replay and share.