Show HN: LLMadness – March Madness Model Evals (llmadness.com)

In a spirited challenge reminiscent of March Madness, the "LLMadness" event has emerged, pitting 15 AI models against each other to predict outcomes for the 2026 Men's Tournament. Each model aims for a perfect bracket, employing a unique round-weighted scoring system with a cost tiebreaker to evaluate their predictions. As participants, notable models include OpenAI's GPT-5 series, Google's Gemini, and Alibaba's Qwen, among others, all competing for accuracy in a high-stakes test of their predictive capabilities. This competition holds significance for the AI/ML community as it explores the limits of current large language models (LLMs) in a dynamic, real-world scenario. Despite the high-profile entries, most models are struggling with zero accuracy, highlighting the challenges inherent in predictions influenced by numerous variables like team performance and historical data. As the tournament unfolds, the outcomes could lead to insights on model performance and efficacy, guiding future advancements in AI predictive modeling and offering a playful yet rigorous platform for evaluation among leading technologies.