I’m Harishankar, creator of tinybrains.dev. Massive LLMs and billion-parameter models get all the hype, but there’s an incredible amount of power, speed, and elegance hidden inside tiny neural networks. tinybrains.dev is a competitive AI arena inspired by classic multi-agent challenges (like the Google AI Challenge: Ants). It’s designed to challenge developers to achieve maximum strategy using minimal parameters and low-latency inference. Why tiny brains? Compute Constraint as a Feature: Instead of burning GPU power on massive models, you are challenged to optimize small, lightweight neural networks (~100MB or less) to make fast, strategic, real-time decisions. Real-time Strategy Simulations: Write your bot, upload your model, and watch your tiny brain command units, gather resources, and defeat opponents in grid-based tactical games. Fair Benchmarks & Low Latency: The arena runs lightweight CPU inference so anyone can build, test, and compete without needing expensive server infrastructure. Who is this for? Reinforcement Learning & AI Enthusiasts looking for a fun benchmark. Developers & Engineers who love competitive coding (Codeforces, LeetCode, Screeps). Students & Educators wanting hands-on multi-agent simulation experience. I’d love for you to test out the arena, submit your first bot, and give your feedback on game balance, API design, or features you’d like to see next! Happy training, and let the best tiny brain win!

tinybrains.dev

Competitive AI arena for small neural networks

Competitive AI arena for small neural networks