Hi everyone,
Remember Topcoder Marathon Matches and Google Code Jam? I built something for people who miss that kind of competition.
https://tinyai.run/ is a programming contest that never ends. You write a chess or shogi engine, submit it, and it competes against other engines around the clock.
The twist: compute is strictly limited.
- Tiny binaries: One WebAssembly (WASI) file, max 4 MiB(, plus up to 8 MiB of data files).
- Deterministic compute budget: 4 billion WASM instructions per move (roughly 100 ms). Not wall-clock time. Every engine gets the same budget regardless of hardware, and games can be replayed and verified.
- Memory has a price: The first 1 MiB is free. Beyond that, using more memory makes each instruction more expensive. At 32 MiB, your effective compute budget drops to roughly 35 ms worth of instructions.
You can't just throw more compute at the problem. You have to write a smarter engine.
How it works:
Build an engine in Rust, C, Zig, or anything that compiles to WASI. Test it locally using the same referee as the server, then register it via plain HTTP.
No accounts. A small proof of work replaces sign-up.
New engines must first pass three qualifying games against a weak house engine (no losses, at least one win). Once qualified, they enter continuous competition and appear on a public leaderboard with ratings and uncertainty estimates.
Starter engines are available for both games. The smallest house engine is just 6.4 KB.
Why I Built This
Frontier AI increasingly depends on massive computational resources.
I wanted to explore the exact opposite.
How much intelligence can you squeeze into a tiny, fixed compute budget?
A place where a clever algorithm can beat a massive model.
Where a single developer with a laptop can compete with anyone.
That's the kind of place I wanted to build.
Prompting is easy. Winning takes engineering.
In an era where a single prompt can generate an entire application, there's still a place where engineering skill makes the difference.
Top comments (0)