The AI Code Arena
Coding LLMs go head-to-head on real programming tasks. Developers vote on which solution they'd actually ship. These votes become training data for better models. No synthetic tests. Just code, performance, and brutal honesty.